DAtum

Blog migration

2006-03-30T20:21:00.000+05:30

So.. I have finally migrated to the blog software I personally like very much - WordPress.
My new blog is at http://sandeep.wordpress.com.
My feedburner feed has not changed, so it should automatically be working with the new site.

Should we emulate the US in education?

2005-12-23T10:26:00.000+05:30

I read this article, which quotes a speech given by NR Narayanamurthy, where he talks about what to do about India's education infrastructure.

I could'nt help but notice, how his view is clouded by the apparent panacea of Indian IT industry. I believe what he says is wrong.

Narayanamurthy talks about the need to liberalise (a weakly clouded reference to privatise education) Indian education infrastructure. Well, if that is implemented exactly as the US does, we would be heading towards a disaster which would be borne by individual families.
I belive that everyone should be entitled to college education - even though she may turn out to be a future beauty queen, rather than a Ph.D. in nuclear physics (although she may play one in the movies!!). College education is not just about getting the skills required to create nanotech microchips, but also about personal empowerment. It is the way that people, and especially women, enter a crucible to hone their opinions, ideas and goals. I believe it to be the very basis of a growing democracy - free of dogmatic opinions shaped by mob behavior and more of individualistic perceptions.
Harvard University has a corpus in excess of 26 billion dollars - while the average student debt has increased from 160 dollars per month to 210 dollars per month.
Something wrong with this picture?
The US Govt is cutting its aid budget which would have otherwise benefited students. But what is the root cause for the govt. having to aid students - profit making colleges.

Let me not be misunderstood - I'm completely in favor of paying professors very,very high salaries to retain them. Hell, I come from a place where I was fortunate to study under some of the best professors in the world. I'm in favor of students shouldering atleast some of the cost of running labs, etc. But lets be honest with each other - Universities and Research bodies make more money off patents and intellectual properties and less off the steady stream of student fees.

But I do not understand the behavior of making more and more money off universities. I understand Harvard, Stanford, etc. use their funds to finance startups, but 25 billion?? Please, let us not fall into the trap that will end up denying a large population of college education, just because Mr. CEO is facing a shortage of code monkeys.

And what about the justification that subsidies must only exist in basic education ? I would rather pay an education cess that goes on to empower a large section of the populace in, say Bihar, to strengthen our society and democracy as a whole. People like me will be found and nurtured, with a countrywide talent search drives and by interacting with quality professors/educators, and less by liberalising our education.

We may have lesser funkier labs than MIT, but I had in my classroom poor students who would shine in their academics and help us fellow students along. That should always be the spirit of education - the Gurukul way ... an inherently Indian way.

How then shall we get the labs that are a must for technological leaps? The same way that the US does them - encourage intellectual property discoveries and patents and let the University share in the profits. But a very important difference is - plough back the profits into education, not an investment bankers portfolio.

Mr. Narayana Murthy ... you are wrong.

The road ahead...

2005-12-01T20:52:00.000+05:30

I have been short of posts these past few weeks - and this is why.
As of December 1'st, I gave up my position as Lead Engineer in Atrenta and am now a Member of Technical Staff at Calypto Design Systems.

I had a fabulous two and half years at Atrenta and I felt it was time to work on something new. I think it will be cool - with its small bohemian location in the heart of Noida and a closely knit group of hardcore developers.

Lets see what rabbit I can pull out of my hat!!

[UPDATE]
I did not, or rather did not have the heart to write a lot, the last time. Atrenta has been very good to me for the last two and a half years that I have been there. I am proud of my project - Constraints and how it has made a difference to the design process.

I have worked with (and made many fabulous friends) across three continents and it was a great experience. Moving to Calypto has always been about the cool stuff that they are doing and how I will have a great time working there.

That's that....

Looking for and Developing leaders

2005-11-17T11:58:00.000+05:30

A comment by A.G.Lafley, CEO of Proctor and Gamble, set me thinking. In my workplace, I have tried to cultivate people, who as part of my team have shown promise.
It is not always the case that I can recognise the gleam. Frequently, people are still in college mode or are not driven - probably as a result of the babu syndrome. They believe that they would eventually get there by means of seniority.
I think it is important to follow a two-step approach:

Shock and awe them with what you can do: you have to prove your capability as a leader, not only in leading and managing, but also in actual execution.

The Talk: This is when you gently but firmly hold a mirror in front of them - show them how they can be even better and really shine.

I think one of the most important qualites to have whilst doing the above is candidness. If a bloke is an ass, call him an ass. But show him how you were an ass once and you fought for what you wanted to become.
It has worked for me. I think, atleast I like to think, that I have groomed people to be better than what they started with and faster than they would have done on their own.
A fine legacy, I believe.

You mess with one of us...you mess with all of us

2005-10-29T20:45:00.000+05:30

Bomb blasts @ Diwali shopping markets, October 29 2005

The resounding images being that of Delhi-ites pulling out people from debris and helping to carry injured on their shoulders.

We will fuck back.

Bernanke's gonna go up in a 'copter

2005-10-24T21:23:00.002+05:30

New York Times announced a few minutes ago that Ben Bernanke would be succeeding Pope Greenspan as the Federal Reserve Chairman.
Bernanke is famous for his helicopter drop governance principle. In his speech in 2002, he has interesting viewpoints on deflationary recession.
Deflation is a side-effect of drop in consumer spending, and therefore demand. In response to falling demand, producers have no choice but to reduce prices to ramp-up sales. This deflation of prices eats at profit and operating margins of producers, thereby affecting the unemployment rate, industrial growth, etc.
In response to deflation, a country's financial regulatory authority usually decreases interest rates - to promote spending and to encourage citizens to look for alternative investment areas.

But the concept that was way too cool was - nominal interest rate hitting zero. In an extreme situation, nominal lending rate might hit zero, but it still might be prohibitively expensive for someone to borrow money. This is if deflation is prevalent across the entire economy causing the repayment to be made in dollars that have a purchasing power much higher than what you borrowed. (!!)
Now, according to conventional wisdom, once nominal interest rate hits zero, there is further scope for the regulatory to do anything but wring their hands in despair. But Bernanke's thesis goes on to the printing-press theory.

By increasing the number of U.S. dollars in circulation, or even by credibly threatening to do so, the U.S. government can also reduce the value of a dollar in terms of goods and services..

Therefore fueling demand. Is this just a theory?
The Conspiracy
In 2003 and the first quarter of 2004, the Japanese govt. printed several trillion yen. This yen was used by the Bank of Japan to purchase US Govt securities at very low prices. Coupled with tax-cuts by the Bush administration, this led to to savings-glut led surplus turning into a spending deficit within a few years. Increased consumer spending in the US caused Asian economies to accelerate, who in turn used their dollars to settle their deficit with Japan.
The circuit is complete.
Is Bernanke's theory sound? I dont know.. but it sure is cool!!

When VC's turn evil

2005-10-23T18:33:00.000+05:30

Bill Burnham blogs about the VC investment process in Datapower and the happy ending.
It was very interesting that sometimes VC's can gang up together and force an entrepreneur to accept funding at their terms or valuations. Even the comments section of Bill's post has one about someone who had to seek funding from out-of-area, to get a better deal.
Problem is, that in case initial investors dont invest in the startup, it is a warning flag for other VC's. This way, a group of VC firms can armtwist a small company into giving it all up.
I'm curious as to how the Datapower guys handled it - did they tell Mobius Capital that they were trying to look for a better deal? What was it that they gave as explanation for not accepting the original investors term sheet?

Explicit construction of class object (not pointer to class object)

2005-10-22T17:48:00.000+05:30

Consider the following piece of code:



#include <stdio.h> 

#include <stdlib.h> 
using namespace std;

class SSS{
    public:
       SSS(){
        printf("Calling constructor\n");
       }
};

typedef struct test{
SSS s1;
}test;

int main(){
    test *t1;
    t1=(test*)malloc( sizeof(test)*1);
    return 0;
}

Does the constructor of the class get called?
Nope.

Then how do you do it. The problem is, how do you explicitly call the constructor of a class object that is defined but not constructed. Usually, class objects are constructed as soon as they are declared. For e.g.

SSS s1;

The above will immediately call the constructor of SSS. However, this is a case where objects are declared, but not constructed. And there is no easy way to construct it now.

First of all, this is bad design. If there is ever a necessity to explicitly declare a class object but construct it only later, then a pointer to a class should be used.
However, this is comes from another problem that I encountered.
Legacy code.
Ah yes. the bane of any code writers existence. While code can almost be thought of a s poetry, this, ladies and gentlemen, is nothing short of blasphemy.

In any case, the solution is not very difficult - placement new operator.
Take a look at the modified code:



#include <stdio.h> 

#include <stdlib.h> 
using namespace std;

class SSS{
    public:
       SSS(){
        printf("Calling constructor\n");
       }
       void* operator new(size_t size, SSS* s2){ //<-- need to overload

        printf("Calling operator new\n");
        return (malloc(sizeof(SSS)*1)); 
       }
       private:
        int n;
};

typedef struct test{
SSS s1;
}test;

int main(){
    test *t1;
    t1=(test*)malloc( sizeof(test)*1);
    new(&t1->s1)SSS;  //<---- how cool is that

    return 0;
}

The key is overloading the new operator and calling the placement new operator in a weird way:

new(&t1->s1)SSS

Again... bad design.

Fly like a bird

2005-10-16T18:45:00.000+05:30

2 hours 25 minutes 30 seconds
That's my ChampionChip electronic time today. 21.097 km in 145 minutes - I missed my target by 20 minutes, but what the hell.
Running is like fighting in the ring - after a while you become concentrated on just the next step, the next punch. It is like meditation, with the body and the mind.
I paced 14 km with an Italian guy. We watched each others pace constantly. And we were cheered on by the multitude (It also helped that my companion used to urge them on.. qite vocally!!). After 15km, it became sort of surreal, I could not feel very well, just knew I had to run. Especially the run down Rajpath... wow.
Never stopped..never broke the pace, even when we had to run up a flyover. When I sprinted the last mile... well, Lance Armstrong helped me along. I could hear Captain Nemo

There I am independant. There I am finally free!

Oh and the cute cheerleaders helped too... I'll blame my excessive endorphins today for being such a shameless flirt.

Music Trance by Lagoona

Marathon training: Week 2

2005-10-12T15:24:00.000+05:30

Could only go upto 10 KM in 55+ mins. Also have a bad case of bleeding due to the shirt chafing my skin.
Note to self: get bandaids for the run.

(Re) Looking at your company

2005-10-08T09:44:00.000+05:30

I discovered this post by Sramana Mitra.
It is about taking a critical look at the problematic child of a VC's investment portfolio.
I however see it another way - I see it as the information, a founder must always have at his/her fingertips. I feel this way after being inspired by another piece by Bob Parsons (founder, GoDaddy).

The whole point? It can be boiled down to :

Everything that is watched improves

Parsons takesa a leaf out of the life of John D Rockefeller - who was famed for his minute knowledge of all things oil. He was probably the only person in Standard Oi, who knew exactly how much it costs to extract, refine and bring to market, a barrel of oil. This, in the days before Excel.
Similarly, out of Sramana's points, I feel a few are the ones that should be worked upon.

Which Market? How big is the Market? Why?

What Value does the Customer see? What Price can you charge based on Perceived Value?

Both of the above points are related to pricing. There is a problem with these points as compared to the others.
For example, resolving conflicts in Sales Channels, How to define Lead Sourcing/Generation/Qualification, etc. are solved problems.
The issue with pricing is, only the founders should be qualified enough to fix a price - they know what their technology is and they know who to sell. Pricing is a problem, that if attempted to be solved by sales guys or code monkeys will result in a F***edCompany.
The other thing that I feel is a tough problem to solve is

What are the key additional hires / timeframe?

No matter how good you are, you cant do everything. However, it is an extremely tough problem to acquire new brains. Especially in a startup, where you may have to pay a lot more than they are actually worth.

measure and watch everything of significance

Another point by Parsons. Sounds paranoid, but as a founder helps you give clear warning signals. For example, in a software company, a very plausible scenario is where a core module is being delayed causing all other modules to be delayed. However, as a founder you would trust your lieutenants to get the job done.
What if it is'nt? What if the lieutenant bit off more than he/she bargained for?
As a founder, I think I should know about it much much before it is beyond salvage. How about a metric to measure the lines of code committed everyday in each branch, in each module. Will this help me to flag warnings?
I dont know, but I think it will.

Delhi Half Marathon Training: Week 1

2005-09-30T16:53:00.000+05:30

Siri Fort: (post tennis session)

Day 1: 5 Km - 21 mins

Day 2: 5 Km - 23 mins

Day 3: 7 Km - 35 mins +

6 eggs + 90 gms whey .

Got to go up to 10 kms next week in 45 mins. Think of Lance. Always think of Lance.

Chapter 71

2005-09-30T09:50:00.000+05:30

71'st chapter of the Tao Te Ching

Not-knowing is true knowledge.
Presuming to know is a disease.
First realize that you are sick;
then you can move toward health.

There is no spoon.

Inspired by Evelyn Rodriguez's post.

Does heapsort make more sense for EDA algorithms ?

2005-09-29T20:50:00.000+05:30

Just a random thought - given that netlist representations of most HDL appear in the form of unbounded trees (parents with greater than 2 children), this can be most easily be represented by list of lists.
Now, if we use a custom memory manager that pre-allocates large chunks, this means that the lists which will be created, be most probably contiguous in memory.. or we can make them contiguous, without loss of great processing power. This in turn means that they are similar to arrays.
Now, heapsort will perform with O(n log(n) ) upper-bounded complexity. But, if the array like structure is enforced, it will be very easy to map it to a similar heap structure (just re-ordering). Which means that heapsort may have better cache-performance than, say quicksort (cos the temporary space required is inside the orginal array heap).
Just a thought.

Prediction markets: real life psychohistory?

2005-09-25T19:44:00.000+05:30

This summer, while I was in Tunis, I read a story by Robert Reed called the Opal Ball. It tells of a time with a sort of Ebay-like marketplace, where instead of goods, predictions about the future are traded. It is the norm to put details on potential dates, career strategies, etc. on the market where they are traded like any other futures and it is only quite infrequently that the Market is wrong. The couple in the story tries to go against the market and get married, however it ends up in a bitter divorce. The infant daughter however is kept away from the Market by the father, perhaps in an effort to let chaos run its course.
Perhaps it is because of Heisenberg. A precept of quantum theory says

that which is directly observed as the act of observation changes the nature of that which is observed

This is not a work of fiction.

Friedrich Hayek is considered the father of predictive market theories. He had a strong belief in the fact that price mechanisms carried more information than just its inherent content - which is to say, price. He always maintained this about price mechanisms

that which is the result of human action but not of human design

Language, for instance, carries more information than its content. The pretty little thing at the bard could have said, "I'll call you tomorrow", but its the eyes which tell you if she really will.
This is the basis of the Iowa Electronic Markets. It gained notoriety in accurately predicting the results of the US presidential elections.
How is it different from a poll? A poll is a snapshot of the current distribution of "consumer" preferences. However, in prediction markets, you introduce competition. Therefore, now you analyze the problem with an intent to win. However, you are also part of the problem that needs to be solved.
When designing prediction markets, the risk free rate and required expected return rate is constrained to be zero. The IEM operates two kind of futures (if we consider election outcomes as the market):

Winner-takes-all: contracts pay 0 or 1 dollars depending on whether the candidate won or lost the election

vote-share futures: contracts pay an amount equal to the fraction of vote-share received by the candidate times $1

Intuitively, the vote-share futures can be seen as a distribution of probability of vote-share (think of it this way,for a given contract price - the vote share - the volume indicates his perceived probability of winning). The winner-takes-all futures are actually a subset of the vote-share futures - you are actually pricing the candidate when his vote share is the highest vote-share. The interesting thing in all this is, you actually factor in other people's expectation of votes, while constructing your own. For me, that seems to be an incremental refinement of knowledge.

What makes this abstruse topic interesting now is the fact that Google is building one. Consider what happens: there are certain coefficients in probability theory that needs to factored in when arriving at a probability decision. These factors are usually computed (and fine-tuned) using past history. Google has the resources to implement a large-scale prediction market. Therefore, it will possess these magic coefficients. In addition to knowing what is happening (aka Google News), it will finally know what is going to happen.
"42" does not seem to be so funny after all.

Ebay + Skype = Semantic Web - Google ?

2005-09-16T23:26:00.000+05:30

There has been widespread consternation and disbelief over the sale of Skype to Ebay for 4.1 billion dollars. Muchos dineros in my opinion. Even Robert Scoble tries to make sense of it all.
But in my opinion, they have all got it wrong.

Before I go any further, let me tell you who founded Skype - Niklas Zennström and Janus Friis. Two guys who also wrote software to share files and gave the recording industry a severe case of ants in their pants. Given that Skype is an "inflection point" (I seem to be quoting Andy Grove a little too much these days)for incumbent telcos, it leads to reason that there were other, more eligible suitors for Skype in the very relevant and related field of telecom voice carriers.
But they gave it up to Ebay.

Now let us look at a very relevant question - Who is the biggest threat to Ebay? It's not rival auction sites, it is not brick-and-mortar liquidation houses, it is not entities like Amazon zShops. It is Google... or whoever among Google, MSN, Yahoo is going to win the search war.
People "google" for stuff, they never "ebay" for anything. Google is now the defacto standard for our collective intelligence. It is Cerberos holding the keys to everything.
It is a very, very trivial matter for Google to setup its own auction house and the scaling up of its business will rival Ebay's in a much shorter time. Given that Google has shown keen interest in setting up a monetary transaction system to rival Paypal ( and given the fact that half the people are fed up with Paypal - check out the Katrina Relief fiasco) and it wont be surprising if they create something simply too good to be true.
All this is setting up stage for a direct confrontation with Ebay. There is no way in hell that Ebay can create a search engine to rival Google.
Now what?
Two pieces of a jigsaw puzzle - Gmail and Skype. How could they possibly be related, you ask? Ahh but sire, therein lies the key. Gmail has set a precedent that could I like to call "non-invasive invasion of privacy". In other words, targetting advertisements related to the actual contents of my very private emails. Is it too difficult to carry this over to voice? Is it too difficult to integrate sophisticated AI on the Skype software to deliver targeted ads? And since Skype already manages to keep your telephone calls private, routing them through your neighbors computer, it is not difficult to add some semblance of privacy to these targeted "advertisments".

The war here will be fought on battlegrounds of technology as well as legal hallways. Currently VOIP does not come under the purview of regulators - which gives Ebay+Skype some breathing space. But in that time, if they can come up with a legal shield that will allow them to do this, it will all but cut off Google from their channels of distribution.
Building up Skype as the world's leading provider of VOIP is part of the package. That will allow Ebay to reach into homes and tap into your conversations. You could be telling your buddy that you were thinking of buying a not-too-expensive pearl necklace for your girlfriend, and suddenly Skype's software queries Ebay for "cheap+pearl+necklace". The results could be delivered via SMS, Email or by a tele-marketer.
While online advertising has a good enough conversion ratio (converting clicks into actual purchases), this would beat the pants of everybody.

In my opinion this is a mighty fine gamble that Ebay has taken. And I'm surprised no one else has speculated this way (Robert Cringely, are you listening??)

Just a brain teaser

2005-09-09T23:06:00.000+05:30

There is a very interesting puzzle which you would be able to find at several places, but I have an interesting non-trivial twist to it:

Part 1: There are 12 balls, one of which is lighter than all the others. Can you find out which ball in 3 weighings?

Divide the 12 balls into 2 groups of six. Weigh to find the lighter six balls. Divide 6 balls into 2 groups of 3. Weigh to find the lighter 3 balls. Weigh two balls - if they are equal, then the third ball is lighter, else it is the lighter of the two.

Part2: Same question as above, except now you dont know if the ball is heavier or lighter.

Divide the 12 balls into 4 groups of three balls each. weigh two groups - thus locate the 6 balls(in two separate groups) which contain the "Ball"; Also note which group goes up and which group goes down in the weighing balance. Weigh one candidate "Ball" group with the known good balls group - this way you find out which group contain the "Ball" - at this point of time, you also know if the group was the "up" group or "down" group, therefore whether the "Ball" was heavier or lighter. Now the three balls can be resolved as above.

My take on this:

Part3: Same question as Part2, except can you do with an initial group of 4+4+4. i.e. in the first weighing, you are forced to weigh 4 vs 4 balls.

I believe this is a non-trivial solution to this and dunno if anyone else has attempted it.
Weigh 2 groups-of-4. If they are both equal, then the "Ball" is in the third group. This can no be made a 6 ball group by adding 2 good balls. The solution now follows Part2 above.
If they are unequal - note which is the "up" group and which is the "down" group. Here's the complicating part:

**** ???? <- let these be the 2 groups of balls
remove 2 balls from group-1 and 1 ball from group-2. So,

** ??? **?
Now there are group-1, group-2, group-3.

Put 2 balls from group-2 into group-1 and pad group-2 with 3 good balls.

**?? ?000 **?
Weigh group-1 and group-2. If the direction of "up" and "down" has changed from previous weighing, then the fault is because of the two ?? balls transferred from group-2 to group-1. If the directions are maintained then the fault is because of the two ** balls in group-1 and ? ball in group-2. Thus now u have 3 balls - **?
If the weighing is equal, then the problem is in group-3, again ending up with 3 balls - **?

Now, we already know that * balls and ? balls belong to one of "up" or "down" group.

From the 3 balls remaining, take one * ball out, and club the remaining * and ? balls together. So,

* *?

Weigh the *? ball with two good balls. If the *? goes in the direction of *, or in the direction of ? ("up" or "down"), we know whose fault is it. If they are both equal, the remaining * ball is the culprit.
The interesting thing is in the last possibility - we will never know if the ball was heavier or lighter!

You heard it here first!!!

Getting too hot to handle?

2005-08-28T11:38:00.000+05:30

Gabe-on-EDA has a guest article by Tets Maniwa, whow writes about the Hot Chips conference at Stanford. Besides the perfunctory presentation on Moore's Law, it seems a focus was on power vs performance this time, especially by Intel. For example, Intel launched the Viiv platform, which is based on media-centric non-form factor devices. It essentially means chipsets which can perform tasks ranging from plain vanilla web surfing to computing intensive gaming, all taking place under the hood of a variety of form factors.

Form factors.

That is the key difference here. For this to be able to accomodate a variety of form factors, it essential to conquer the power problem.
Enter William Holt and Intel.

It has been common knowledge that leakage power and interconnect losses are dominating current chip manufacturing technology generations. From a manufacturing point of view, they are trying new and funky things like high-K dielectric substrates and the new horse in the fab stable (in my opinion) - silicon on nothing.
This would also cause changes in design methodologies - especially at routing (maybe a problem that companies like Teraroute could solve) and in timing closure. For example, as Tets Maniwa points out, designing with multiple voltage islands is not a simple task. Current design methodologies are limited because of the extensive use of wireload models - these may be inaccurate in the face of dramatic on-chip variation (as is the case with voltage islands). This may require multiple respins of place and route to achieve accurate enough back-annotation of interconnect delays and suitable timing closure.

All in all, its an interesting world we live in.

File and directory check in Makefile

2005-08-24T21:53:00.000+05:30

I needed to do some conditional target-based execution via my makefiles. To make them truly generic, I needed to deduce directory structure that I was sitting in.
One of the things I had to do was to check for existence of a file or directory. Now since Makefiles have a different syntax than ordinary shell, compounded by the problem of interfacing with the *nix shell, it took me some time to figure out/learn how to do this.

The syntax (for GNU make) is

ifeq (exists, $(shell [ -d directory_name ] ) && echo exists )
$(warning hi... I did it)
endif

At first I tried to do

ifeq (1, $(shell [ -d directory_name ] ) )
$(warning hi... I did it)
endif

but this did'nt work, because the shell interface does not return a numeric status (atleast I think so). But in the syntax that works, by the power of logic, the "echo exists" will take place if and only if the "shell" command works.
Ergo, I can check for directories.

whew!!

Is EDA at its nadir?

2005-08-19T21:42:00.000+05:30

Sramana Mitra attempts to make a case for consolidation in the EDA industry.
Very interesting actually, especially the part about

... Likely, this will adjust some of the structural disfunctions of the industry..

and from another post

Such a market cannot support 3 major players...

Well well well... Let me put it this way:

Sramana's entire theory of the EDA industry and its consolidation arises from a basic numbers calculation - decline in Q1'05 revenue (989 mil. USD) vs Q1'04 (995 mil. USD).
How this reasoning coupled with the deliciously abstract theory, of this size being disproportional to three major players, is what drives the logic.
Let us analyze these numbers even further. The average productivity per person in the EDA industry is approximately 197000 dollars (derived from the EDAC MSS report). Now going by Culpepper's employee productivity report, that would be well within the top 2-3 industry segments in the technology arena.
The problem is the revenue margins for the EDA industry. Gabe Moretti had an article up some time back, and so was my reply. The issue is that there are far too many problems to solve and sufficient venture capital available for these problems, that this industry is moving at a break-neck speed. For example, look at Sierra Design Automation and its Pinnacle flow system. It is trying to shake the industry. The same is true for Magma (lawsuit non-withstanding).
It would be a happy and sepia tinted world with only three companies in the EDA arena, but then I suggest to go and talk to design companies - their frustrations with existing design flows and need to move to something much more intuitive. The other day there was an interesting article about 45 nm technology and its challenges. Who is going to figure out how to handle this at the design level, rather than at the fab level? Is it going to be the big three, or some startup? Will consolidation make it easier or harder.

And while we are on the topic of acquisitions, let us look at these opinions by Wally Rhines, CEO Mentor. Standalone IP vendors lack the application knowledge as well as the support and application R & D infrastructure (read: engineers skilled to debug problems with glue-logic and back-end design flow). IP does not always work.
And here's the biggest bummer of all - ESL will work..maybe not now..maybe not next year, but it will work in a few years. It is a logical extension of compiler technology. Its the same thing that happened with transition from Assembly Language to C/C++ to PHP - how good is your underlying compiler technology.

The EDA industry is riding the fast train of innovation, fueling itself with newer and more challenging problems. And there is a lot of steam left!!

Happy Independence day....

2005-08-15T12:07:00.000+05:30

Celebrate by having kababs and curry and patiala pegs of scotch!!

Intel's strategic inflection point

2005-08-12T21:11:00.000+05:30

The term "strategic inflection point" was coined by Andy Grove, CEO extraordinaire, in his book Only The Paranoid Survive. The phrase is used to term the oft missed jitter in a company's business curve, which actually turns out to be what makes or breaks the company.
It is my opinion that Intel faces that again.
Let us see what happened the last time. Circa 1980's : Intel's main product line - memory modules - had been all but cleaned up the Japanese. It was a turning point, to invest money and Ph.D's behind its mainline memory business, or to drop everything, turn around and try making processors. In retrospective, it was the most amazing decision, because it put them ahead in a curve which was just about to rise. But if I try to think as Gordon Moore, Bob Noyce or Andy Grove would have thought, I would have been faced with the same crossroads - why abandon a perfectly good product, which we can almost win and rather turn around? The answer was in the world around them - an increase in necessity to compute, primarily by businesses, was something that was only going to grow.

Circa 2005: Computing power is commodity. A record of the fastest supercomputer is made and broken in less than half a year. It again brings us to the need to look at the world around us. There have been news of Intel first threatening to and then actually not backing down from low-end desktop markets. However, that is not what counts - the world today revolves around what I call Media Delivery Platforms. ITunes/IPod/Mac Mini is one such platform.
The Cell-Processor/PlayStation3 promises to be another, as is XBox/XBoxLive. More and more business applications run on platforms and operating systems (Linux + Xen virtualisation), that allows multiple operating system, multiple software running parallely and completely compartmentalised. It is in Media Delivery Platforms, where people can be locked in.
Is this technologically possible?
Certainly. Technologies like Palladium enforce Digital Rights Management (DRM) at the hardware level. But to ask people to buy into your products means that your products are easy available and serviceable at the corner store. Which means it would be a really bad idea to move out of the low-end desktop market. Because it is these platforms which are tried, tested, and are cheap enough to penetrate the kitchentop markets.
It is here that the war for the next inflection curve will be fought.

Mumbai meri jaan

2005-07-30T17:08:00.000+05:30

I love my Mumbai and the people. They came and they fought back and they helped other people.
I spent four years in IIT-Bombay, but even after living away I realised, you can never take the Mumbai out of a man.

Weird TCL behavior on Cygwin/MinGW

2005-07-26T21:37:00.000+05:30

A C++ program compiled with TCL stubs, which performs just fine on Linux, just seems to go out without a sound on Cygwin. TCL_LIBRARY was set properly (also tried the

"export TCL_LIBRARY=$(cygpath -w /usr/share/tcl8.4)"

hack). But the execution just seems to vanish after a call to Tcl_Init(). Now internally, Tcl_Init parses a whole load of stuff, starting from init.tcl.
Also tried a weird hack that changes the "tclPreInitScript" variable to help set the $tcl_library variable, et al. But nothing helped.
On debugging the executable in gdb, the call to Tcl_Init returns a segfault, which is not visible on a normal run.
The solution to this mystery is a simple call to

"Tcl_FindExecutable(argv[0])"

, which sets up some internal variables for use by tcl. Now since in cygwin there are two paths (the /usr type path and the /cygdrive/usr path) for all resources, this helps initialise everything correctly. I almost did'nt try this solution.
But it works!!!

RTL Signoff - When will it happen?

2005-07-25T21:26:00.000+05:30

A post by Gabe Moretti set me thinking. Exactly how critical is RTL handoff going to be ? By critical, I do not mean to the designers - but rather to the upper management of the semiconductor companies, who are the only ones with clout to influence a radical change in the revenue model.
Design for Manufacturing is not a new keyword. But I think this keyword now has competition - One of which could be Platform Based Design and the other could be design methodologies like the RapidChip environment. This essentially reduces the argument for profit sharing by foundries and semiconductor cos alike. Atleast this is the bargaining chip that could be used by them to push for RTL Signoff as just any other methodology and being priced similarly.
There was also an interesting article by Aycinena on how foundries may end up acquiring EDA companies, for much the same reasons that Gabe puts forward. However, I feel that, given the step-fatherly treatment that semicos mete out to their EDA underlings, it is far more probable that EDA tools will remain a commodity item.
Then who will solve the problem of RTL signoff, if indeed it needs to be solved (refer to alternative methodologies above) ? Well Gabe gives the answer himself - VC's and startups. Oh and if I may point out an earlier article by ~~Gabe~~ Dieter Ernst , it may well be done by people having Tandoori chicken for lunch!

Venture Capitalists and the "manager-ese" language

2005-07-21T13:02:00.000+05:30

A great post by Narasimha Chari, underlines something that I had suspected all along - the standard VC questionaire on "barriers-to-entry", "competition-space" and other manager-ese lingo is plain marketing. If you can maintain a straight face and confidence when faced with these, then it can be handled. Ultimately, no one knows if a venture will succeed or fail (who would have bet on iPod and how many did'nt on Webvan), but do you have the passion, hustle (a-ha!!) and resiliency to.

A Sardarji in New York

2005-07-18T21:02:00.000+05:30

As I have written earlier, the US is looking towards India as a long term strategic and economic partner - both as a key foothold in the labor-rich Asian arena, as well as in its fight against "terrorism" (the double quotes being entirely intentional). As Prime Minister Manmohan Singh (or Moneymoan Singh as he is so fondly known!!) commences his tour of the US, a question lingers - how much armtwisting will the other do?
The Bush govt. will go to far lengths to gain a security foothold in India. It has already got the Indian govt. to agree to FBI presence on Indian soil - with all its anomalies about human rights. However, I am betting that PM Singh will wrangle his fair pound of flesh.
India has already made it clear that it wishes for the H1-B quota to be tripled. Though it might appear we are encouraging..nay..actively fighting for brain drain, it has become increasingly apparent that the pull of the paratha(pizza nonwithstanding) is too great for most Indians, who come back with more skills than they left with. I believe this will be one issue that will be put on the table during talks with the US Govt.
What I hope for, but do not have high hopes for, is for India to push for more liberal laws in the generics pharma sector. India is the largest manufacturers of generics in the world and the most innovative. Since Indian manufacturers jumped into the fray, the average cost for anti-retroviral drugs nosedived from USD 10000 to USD 250, in little over 4 years. Brazil and several African countries source their drugs from India. The TRIPS agreement enforces these rights by clamping down on generics production. I believe there is a way out of this by having an Eastern Economy Edition type of system to provide differential pricing (enforced by providing appropriate pricing to citizens only).
The hot topic of the day will of course be India and the Security Council. However, considering that US is not ready to antagonise both China and Pakistan, there seems little chance of progress. However, since US has already dangled a 10 year defence partnership with India, I believe that 10, Janpath will not push for more. As India builds INS Kadamba into a major naval force, it will affect the security of not just India, but also allies like Japan who depend on maritime shipping through the Arabian straits.
And as Pavitr Prabhakar would say, "with great powercomes great responsibility" - lets see whether India can exert her power or discharge her responsibility.

Ha..Ha..Ha

2005-07-11T15:59:00.000+05:30

This probably wont be understood by any Muggle who has'nt read Ayn Rand, but here's The 25 Most Inappropriate Things An Objectivist Can Say During Sex
Cool..

Quest for Fool's Gold: India and the Security Council

2005-07-03T18:19:00.000+05:30

With events in India's diplomatic circles reaching a silent crescendo (seems like an oxymoron..but that's how cloak-and-dagger it is), India's quest for a permanent seat at the security council has reached a new sense of urgency.

Is this all that necessary?

For one, since the United States crippled the status of the new members, by refusing to grant them the veto privelege, the G-4 (as India, Japan, Brazil and Germany are known) have accepted fool's gold. As it stands now, of the G-4 countries India is the one country that can provide a security, economic and political counterbalance to the Chinese influence in the South-Asian region. Japan - being dependent on the US for its security, Germany - which had better resolve the EU constitutional crisis before flailing elsewhere, Brazil - a relative newcomer to global political innuendoes are not going to stand as an effective counterbalance to the sphere of Chinese influence.

Though China recognises that, in light of increasing Indo-US ties (helped in no small part by the highly influential expatriate Indian caucus in the US), it had better maintain growing trade and economic ties with India. Similarly, we recognise the need to forge strong economic bonds with China and the ASEAN.

As it stands, India is the country which has consistently opposed the United States on issues that appear to be skewed - be it the CTBT, patent reforms and Iraq. As the only democratic rising power in the Asian region, it has the potential (as yet to be exercised - a case in note would be our silence to the Taiwan issue, with whom we still do not diplomatic ties) to chill the fire of the Dragon's breath.

However, it is moot whether or not a permanent seat can help in doing this. A permanent seat is a recognition of power, rather than a seal of approval to exercise it. Can India, which has not opposed China on many issues (other than the venerable Dalai Lama), can do what it could not earlier? I believe that a permanent seat without the veto option is just a permanent seat-warmer with the option to talk out loud.

For me, I think the Indian govt. should pursue the idea of the Ruan - a unified currency for south-asia along with creation of a Free Trade Area (a'la EU). That would be much more effective than a frikking permanent seat.

Rise and Shine

2005-06-30T13:24:00.000+05:30

I have been promoted - to Lead Engineer.

Free beer today !!

Attitudes towards leadership : Asia, America Europe

2005-06-30T10:17:00.000+05:30

An interesting article, over at Harvard Business School - Working Knowledge got me thinking.. or rather reminiscing. The author, Prof. Quinn Mills, talks about differences and/or similarities in management and leadership styles between different continents, and therefore different cultures.
The three predominant styles of management come from Asia, America and Europe. But perhaps this is not illustrative enough. Let me rephrase it - the three pre-dominant styles of management are the Hong-Kong/Japanese, United States and Franco-German . Perhaps, I am not qualified (or as I like to put it, polluted) enough to pepper my thought with manager-ese jargon, but based on my working experience on three continents, I would say this would be the case.
But before going further, let me clarify what does not constitute an individual style of management. For example India - we do not have a style of management that is entirely our own, unless of course you would like to classify the bania style of management (what with its glib way of talking into and out of situations). I believe that the pre-liberal India had a a leadership which is not too uncommon across the world. This is what we call babu-dom and the rest of the world calls bureaucratic. In other words, there was no leadership. Increasingly, we have been influenced by the American way of working, not in small part because a large number of expatriates worked in America and among other things, built the frikking Silicon Valley.
On the other end of Asia, there is the Hong Kong-Japanese way of leadership. Japan has a style of leadership (Howard Stringer of Sony non-withstanding) that is to a large part influenced by its feudalistic society. I, of course, refer to the Zaibatsu and Keiretsu systems. I believe that the Hong-Kong/China systems of large businesses have increasingly followed this approach. With it comes a highly stratified system of management (what I believe Prof. Mills calls Directive Leadership). However, this brings in itself a sufficient lack of adaptability. I believe the Indian leadership, with its touch of Asian discretion and mixing business and personal life, along with sufficient individualism (which leads to what the good professor calls Empowering leadership) bridges the continents beautifully.
However, I take exception to the fact that political ties are less common in Asian businesses. If anything, they are more so. This is usually because the corporate process is often much more opaque in Asia, which leads executives to seek political intervention on several issues. Take for example the issue about Reliance Industries and the CDMA/GSM war in India. There was heavy multi-party political influences.
Which leads the remaining style of leadership - the Franco-German. A style which you are born into. This style is getting increasingly threatened by the scales of efficiency and production driven management. It is rare to see European companies laying of workers. However, ST Microelectronics, a French-Italian company is cutting jobs in Europe while increasing positions in Asia. If anything else, this is a change in the thinking of European leadership who view Asia as a threat to their way of privelege.
I think the world is adapting to the Asian-American way of leadership, with its peculiar brand of passion towards work, late night pizzas and an Asian sense of respect and honor towards work and peers.
Tikka burgers anyone?

Why I cant be a good industrialist...yet

2005-06-28T20:59:00.000+05:30

This comes fresh off the horse's mouth - my manager.

Due diligence - Get an attitude about mastering whatever you are working on. If you are buying a piece of land, understand the direction of wind, water table beneath your feet, the number of cows in the area. Good businessmen cultivate an attitude to do their homework thoroughly before doing anything. Risk taking comes much later.

Learn to gauge people - At a glance understand what his strengths and weaknesses are. A consistent attempt to understand people must be made ..what makes them tick.

An attitude towards money - make every rupee count. There must be an attitude to push each rupee to its maximum. It is OK if you lose everything, but there cannot be a situation where you dont know why you lost that money.

Failure is not an option. Always, always, always get the job done.

These wont hold me back. Not for too fucking long.

From startup to... sundown: comments on viral marketing

2005-06-27T22:57:00.000+05:30

There's a great post by the kind people of Drake Investing. It is a highly condensed, in-a-nutshell view of what goes through an angel investor's mind throughout a pitch.
A common vein that I see in many VC/angel communiques is the keyword 'viral'. It is one of the two oft-abused jargon (the other being 'long-tail'). I have written before about how things get viral. It is a common perception that exclusivity breeds popularity. Kind of a oxymoron...but is there any truth in it?

This article in Always-On may provide a clue. The first step towards generating a viral buzz is to harness the power of the collective intelligence - the early-movers who have an ephemera of interest long enough to comment on it, write a passing mention to it. Identifying and targeting these people would be the starting point towards generating a word-of-mouth momentum.
For example consider the 'Pirillo Effect' - the founder of Lockergnome is famous for searching out mentions of his name and posting comments (earlier days I hope!!). It is entirely an effort to try and pinpoint the early-movers who would later help him to become a blogosphere celebrity. And guess what, it worked.
Though such marketing tactics work well (and are well desired by investors) for eyeball-retentive businesses, they might be off target for more mainstream businesses. For example, while it is quite understandable that Gmail generate a buzz by maintaining the facade of exclusivity to penetrate the already overcrowded email arena, I do not believe Salesforce.com would do as well employing similar tactics. That is because the channels of intelligence that potential customers of Salesforce subscribe to, are not mainstream. Instead they depend on single, focused sources of information, a.k.a Engadget, Extremetech, etc. Listening to feedback from such sources, may be passed off as listening to the community.
This was employed by Microsoft RSS team when theymodified their spec in response to Phil Ringnalda - a move which earned them accolades.
Of course, the Scobleizer helped too... but more on that next time.

Function pointers: number of parameters mismatch

2005-06-20T22:09:00.001+05:30

Here is a strange quirk of gcc - It ignores number-of-parameter mismatch if we use function pointers, if and only if we pass more arguments than necessary.
Consider the following piece of code



#include <stdio.h> 

typedef int (*myFunc)(int a, int b);
int func1(int a){
        printf("In func1 \n");
}

int func2(int a, int b){
        printf("In func2 \n");
}

void executeFunc(myFunc someFunc ){
//void executeFunc(void* someFunc ){
        myFunc actualFunc=someFunc;
        actualFunc(1,2);
        //actualFunc(1,2);

}

main(){
  executeFunc((myFunc)func1);

}

As you can see, I am actually calling "func1", which takes one argument. However,I am actually passing two arguments. The code works fine.
Strangely enough, the other way does'nt work - i.e.



#include <stdio.h> 

typedef int (*myFunc)(int a, int b);
int func1(int a){
        printf("In func1 \n");
}

int func2(int a, int b){
        printf("In func2 \n");
}

void executeFunc(myFunc someFunc ){
//void executeFunc(void* someFunc ){
        myFunc actualFunc=someFunc;
        actualFunc(1,2);
        //actualFunc(1);

}

main(){
  executeFunc((myFunc)func2);

}

The above piece of code calls "func2", which actually takes two arguments with only one parameter. Compilation fails in this case.
What is more puzzling is that, in case I had called "func1" directly (without going through all the function-pointers thing), it gives a compilation error of (too many arguments). I can deduce that gcc is doing checking w.r.t the "typedef" in the beginning, but given that ordinarily a function cannot be called with too many arguments, should'nt this be disallowed?
6.3.2.2 of C89 (which correspondsto 6.5.2.2 of C99) states that:

If the expression that denotes the called function has a type that includes a prototype, the number of arguments shall agree with the number of parameters

Which brings me to another "feature" of C vs C++. Consider the following function:



#include <stdio.h> 

int func3(int a){
return 1;

}

main(){
 func3(1,2,3);
}

This gives a compilation error with both "gcc" and "g++" about too many arguments.
However, consider the following piece of code:



#include <stdio.h> 

int func3(){
return 1;

}

main(){
 func3(1,2,3);
}

This gives no error with "gcc", but an error with "g++". Apparently, in c++, a function declared with no arguments is taken as void, whereas in c, it is simply unknown.

Does gcc have a soft-spot for function parameters with too many arguments?

Only time or maybe Raymond Chen can tell.

People in EDA can cook!!!

2005-06-13T12:12:00.000+05:30

A great site - EDA Confidential, written by Peggy Aycinena (editor of EDA Nation).
Today's special - may I recommend the EDA Confidential Cook-off, served with a side of Letters to the Editor.

Constructors vs function declarations...and Koenig lookups

2005-06-12T01:27:00.000+05:30

A very curious incident happened with compilation today.
"but nothing happened...."

That was the curious incident.

Consider the piece of code below:



#include <iostream> 
using namespace std;
class A{
   public:
         A() {_i=5;}
          ~A(){cout<<"Destroying A"<<endl;}
       int _i;
};
class B: public  A{
 public:   
   B() {}
   ~B(){cout<<"Destroying B"<<endl;}
 
};

main(){
  A a( B() );
  cout<<a._i;
}

Now normally, I would think that everything is all right. My expected screen output is:

"Destroying A"
"Destroying B"
"Destroying A" (its inherited)

However, there is no screen output.
To further investigate, let us try to access a member variable "_i" of A.
The compiler squeals:
>>constructor.cpp: In function 'int main()'
>>constructor.cpp: error: request for member '_i' in 'a', which is of non-class type 'A()(B(*)())

"non-class"?? Very strange. What's even more strange is the way that the constructor parameters to A are interpreted - as pointer to functions.

Now, we know for a fact that STL uses functors, so let's try something like this with them.


#include <iostream> 
#include <vector> 
#include <algorithm> 
using namespace std;
class Functor1{
public:
   void  operator ()(int i){ }
};
main(){
vector<int> vv;

for_each(vv.begin(), vv.end(), Functor1());

}

STL has no problem.

Actually, in the first case, C++ interprets A a() as local declaration of a function 'a' with return type A. One of a long threads in the "comp.lang.c++.moderated" newsgroups speaks about this dichotomy.

among one of the workarounds posted is this :


  //A a=A( B() );

Which brings us to the motivations behind programmers declaring functions in local scope. One of the possible reason is to hide an overloaded function declaration. For example, in the following example, the local function declaration overrides the global function declaration:


#include <iostream> 

void f(int i) { std::cout << "int=" << i << '\n'; }
void f(short s) { std::cout << "short=" << s << '\n';}

int main()
{
    f(1); // calls f(int)
    void f(short);
    f(1); // fooled you!

}

However this can be taken care of, by using Koenig lookup - which works only if there are namespaces or classes to be resolved (I dont think it works for primitive types).


#include <iostream> 

class X{};
void f(X x, int i) { std::cout << "int=" << i << '\n'; }
void f(X x, short s) { std::cout << "short=" << s << '\n';}
int main()
{
    X x1;

    f(x1,1); // calls f(int)
    void f(X x, short);
    f(x1,1); // fooled you!

}

Thaks to PK for this !!!

Lead generation and definition

2005-06-07T10:34:00.000+05:30

Great post by Susan Getgood about lead generation and definition.
I think this is the primary point of mud-slinging between sales and marketing. In addition to what Susan says, Brian Carroll writes brilliantly about a universal lead definition.

How to become a VC....

2005-05-27T21:08:00.000+05:30

This just had to be backlinked - there's a great post up by Seth Levine, resident VC at Mobius Capital, about how to be a VC.
Maybe I should show my boss this !!

[Business] Survival of the dumbest

2005-05-22T22:01:00.000+05:30

There is a very interesting story in today's New York Times' Online Edition. Titled "So You Want To Be a Venture Capitalist". It accounts for the travails of a few high tech druids who could not make it in oh-so-no-brainer(?) world of venture capitalism. Which brings me to the topic - Is the corporate world, an arena for the survival of the dumbest?

Take for instance take this story about Google. It describes how the angel investors did some Bollywood-style manipulating to bring about the marriage of two VC's and how the uptight founders had to be loosened up by a very elaborate song and dance routine. But the best part...

“I think John Doerr would say Bill Campbell saved Google,” says Kleiner partner Will Hearst. “He coached Eric on what it means to be a CEO—not the CEO of Novell but of a company like Google. He taught Eric it’s a lot like being a janitor: There’s a lot of shit you have to do.

As my second exhibit, I'd like to present to the jury this article about strategy vs execution. The excellent Fast Company article details the essence of success -

When a senior executive weighs a decision to risk tens of millions on an experiment, surely it is comforting to have an outside expert verify the strategy is sound. But regardless of the qualifications of the outsider, the strategy cannot be taken as gospel. There are too many uncertain factors that nobody can resolve. Even the best strategy is only a hypothesis. No matter who developed it, it must be assumed wrong.

What does any of the above have anything to do with VC's ? Simple - the kind of VC's who will eventually survive are the kind of dumb, paranoid investment bankers who will probably never invest in me at all. Because they will have their 2.5X return and are happy with it. Who wants to be dumb enough to fund the next Google.

However, as a counterpoint, Jeff Nolan has this to say
about picking you VC - a financial investor is a good door opener. However the best advice that he gives is about having a checklist to choosing your VC (How often do they attend board meetings, do they have a track record of introducing you to other deals, etc.).
And now this NYT article. It looks to me that all we are left with is Hobson's choice. What the article does is paint a Quixotic picture of the VC arena, with every ne'er-do-well jumpin' to be a VC.
But maybe they are right. One of the brightest people I know, Rajeev Jain, who is now one of the youngest consultants in Oracle India says this about consultants vs technical leads - "the technical leads only know what is cool and what is cooler, but the consultant knows what will actually be sold.". Maybe having a boring banker for a VC is not so bad after all.

... sign of poor execution was that the leaders never questioned the strategy and never revised it..

Th jury's still out on this one.

[Business] Recovery from cataclysms and exchange rate regime: any dependence?

2005-05-13T23:03:00.000+05:30

Very recently global markets were in upheaval over speculations that the Chinese forex rate would be allowed to move freely against the dollar. In a previous post I had looked at the Zen relationship between China and the rest of the world.
Indian markets went into a tizzy recently over rumors of Chinese revaluation. This was primarily due to bets that the Indian govt. would also allow a higher freedom for the rupee as well as excess supply of dollars - which led businesses to hedge the rupee against rises.
Which led me to think - just how important are exchange rate regimes?

There is a recent working paper by Rodney Ramcharan of the IMF where he explores the effect of exchange rates on the ability of a country to recover from natural disasters, like earthquakes and the like. I especially liked the figure on page 13 which showed that in developed economies, disasters had no impact on investment as a percentage of GDP.
I think this in itself is flawed. Rather than base it on percentage points, I wonder what actual dollar figures are. Any investment in a post-disaster economy (the author cites Bangladesh investing in farm equipment, for e.g.), involves a static cost which would be more or less fixed in terms of cost of living. If we take the Big Mac Exchange Rate, a certain number of burgers would be required in each country to help with food supplies, rebuild roads , etc (That no one in India would eat burgers after an earthquake is moot). This cost measured as a percentage of GDP would figure significant in developing rather than developed economies. This the author himself puts forward in the beginning (page 4) as

In advanced economies,
there is no significant investment response regardless of the regime type. These results are
robust to various modifications, and underscore the importance of the exchange rate in
managing real shocks.

This to me sounds like growth, investment , etc. rate depends on the money remaining in your coffers , above everything else. Consider Philippines for example - it is an flexible exchange rate economy. Now, in case of a natural disaster striking, its growth is expected to rebound much faster than fixed rate regimes (page 31). While the IMF guy says that this is due to higher exchange rates, he fails to say why. Allow me to propose one - full capital account convertibility. That's right, let the brickbats come. Insurance payments and claims, post-disaster, will be made from international coffers to a larger extent. Therefore exchange rates rise.
So either I'm wrong (which is very possible), or the data is just a-priori (which is also entirely possible...anyone give me odds?).

[Software, c++] Separate compilation of templates: Template separation and export

2005-05-08T09:57:00.001+05:30

A recent cause of inexplicable flummoxing and gnashing of teeth was a piece of code like the one below.
Look at 1.cpp - pretty straightforward. An instance 'i1' is instantiated. A function 'myfunc' is called, which returns a pointer to an object.
'SSS1' and 'dummy' are declared in 11.h. Now the templatized function is declared in '13.h' and defined in '13.cpp'. Pretty simple, huh? I agree, there is no problem when compiling this piece of code like this:

g++ 1.cpp 13.cpp -I.

But now consider what would happen if I compile them as:

g++ -c 1.cpp -I.
g++ -c 13.cpp -I.
g++ 1.o 13.o

In this case, the linker, cribs about not finding dummy* myfunc(SSS1<dummy>& a);
This happens because (as I found out recently) a long standing argument in the c++ community on how to separate template definition and declaration. Currently, only one compiler - Comeau C++- provides some sort of proprietary support for separation (via the 'export' keyword). As of GCC-3.4.2, this is not implemented.
The problem occurs because, the compiler/linker uses some sort of optimisation techniques to remove multiple includes of templates and template definition code. when 13.cpp was compiled separately, it threw out basically the whole code, because those templates were not being used. Therefore, at link time, the linker is unable to instantiate the templates used in the function "myfunc" .
The hack (sic) that is appearing to be working is to define the usage in 13.cpp itself.


//dummy* myfunc(SSS1<dummy>& a);
//template <> dummy* myfunc(SSS1<dummy>& a);
template dummy* myfunc(SSS1<dummy>& a);

Notice the addition of the "template" keyword before the definition (the commented line is what did'nt work). Somehow it is important to provide that keyword. Even "template <> " doesn't work.
1.cpp



#include <iostream> 
#include "11.h" 
#include "13.h" 

main(){
  SSS1<dummy> i1;
  dummy* d=myfunc(i1);
}

11.h



template <class T>
class SSS1{
  private:
  public:
   T* member;
     SSS1(){
     member = new T;
     };
};

class dummy{          
};

13.h


template<class T>
T* myfunc(SSS1<T>& obj);
/*{
  return obj.member;
}*/

13.cpp


#include "11.h" 
#include "12.h" 
#include "13.h" 
template<class T>
T* myfunc(SSS1<T>& obj)
{
  return obj.member;
}

[Business, China] Yin and Yuan

2005-05-01T18:47:00.000+05:30

On April 30, the Chinese yuan went out of its government mandated trading range. Currency markets around the worl braced for impact.
What the hell does that mean?

China, like India has always believed that a free floating currency is not advisable till the country's industrial and knowledge base is mature enough to act as shock absorbers. In India's case, the rupee is allowed a little more variation than what China allows its yuan. For example, India's Reserve Bank buys, billions of dollars every day to stem the fall of dollar vs the rupee (This data is maintained by the Bank of International Settlements). The rupee is pegged not against the dollar, but against a basket of international currencies.
The Yuan on the other hand is pegged against the dollar at approximately 8.2760 to 8.2800 yuan/dollar. As mentioned previously, this figure is maintained by the Chinese central bank by buying up dollars. On the 30'th, this exchange rate went to 8.2700 per U.S. dollar, making the yuan much more expensive.

The American Federal Reserve and the G-7 nations are not happy. This is because China and India are the biggest foreign markets for investment. Now as more and more FDI (foreign direct investment) flow in, the central banks buy more and more foreign exchange. This foreign exchange is now re-invested in the US in Treasury backed US Govt. bonds. As a result of this massive investment, the rate of returns decrease for everyone holding the same family of bonds. Now several G-7 countries hold massive amounts of US Treasury bonds and they are not happy about the decrease in the rate of interest.

India, like China, is struggling with a system to maintain control even in crises. We certainly dont want another Asian meltdown. Very recently, the Indian Finance Ministry toughened the path to borrow dollars on the international market. This is part of its effort to reduce FDI. This has not gone well with small Indian businesses who are justifiably wary of any governmental control over business. Therefore the government is now trying to reduce customs duty to decrease the supply of dollars in the country.

But one thing is for certain, India cannot remove the peg on the rupee unless it is assured the yuan will behave the same. It is estimated that the yuan is undervalued by almost 40 %. We buy education, technology and iPods. If the US govt. makes it difficult for Indian students to go US univs. or the Indian govt. makes it difficult to import laptops, this situation will be gridlocked. But the problem is, atleast the government feels, that opening up import trade will harm local manufacturers who have not matured to global economies of scale.
Is that true - only time and China will tell.

UPDATE: Traders are buying up forward contracts of yuan (i.e. a piece of paper that allows them to buy currency at a value fixed now, often by offering the seller a premium)which shows that they expect the yuan to trade upto 5.8 % higher in a year.

[business] Kleiner Perkins and the lost art of the start

2005-04-25T22:28:00.000+05:30

This week, two different stories came into focus - seemingly unrelated, but for me having a slightly deeper implication.
Paul Kedrosky writes about a NVCA report which discusses the investment profiles for venture funds in Q1-2005. The interesting part was that first-time venture funds are at an all-time low in raising funds. First-timers comprise just 12% of all venture funds in the industry.
This sort of worries me - because this figure has gone down from approximately 40 %. Paul has justified that this is because focus is on quality and demonstrable returns. I am worried that this is a general indication of trepidation in the venture capital industry. New funds like Octopus Management's "Eclipse" won the Investor All Star awards in Dec. 2004, but clearly performances like these have not swayed investor attitudes.
Attitudes like this percolates down to the fund itself, in its search for ventures. As Clayton Christensen puts it - "resource dependence", the majority of a company's resources go to satisfy its biggest customers/investors. Does that mean that I cannot do anything, unless I know a John Doerr??
On the other hand, John Doerr himself is hiring "gurus" in their management. Randy Komisar is, among other things, a professor at Stanford. What this means, he is in close contact with brilliant people, none of whom know how to write a truly good business plan, but will have a great Google inside them. Will there be another Bill Campbell to coach them?

[timinganalysis] Why virtual clocks?

2005-04-25T21:53:00.000+05:30

This is something that I learned from my great manager today. Why is there even a need for virtual clocks in timing analysis?

First the funda - virtual clocks are used to set timing-exceptions between two clocks on a particular timing path. There, that's all there is to it.
Consider, two flops arriving at a output port (say out1), through some combo logic. Therefore, there exists, two timing paths terminating at the same output port. Assume that both flops are clocked by two different clocks (C1 and C3). Therefore, it is mandatory that the designer set two output-delays at the output port - one each w.r.t each clock.
Now, while the tool is doing STA, it is likely that it would consider the flop (of clock C1), going through the combo-logic and terminating at out1. However, it will consider two scenarios - timing w.r.t the output-delay set with C1 and w.r.t the output-delay set with C2. Nothing wrong with this scenario, except that it assumes that a signal originating at C1 is going to go through out1 and end up, on the other side, on a flop clocked by C2.
Suppose, this does not happen - as is often the case. So how do we handle this? A simple answer would be to use false-paths. So we put a false-path from C1 -> C2 and from C2 -> C1. Cool. Except, what if somewhere else in the design , C1 and C2 were actually communicating with each other. This false-path would cause those other timing-paths to be ignored by the tool.
Not good.
Enter virtual clocks.
Let us create two clocks C3 and C4. C3 has the timing characteristics of C1 and C4 has the timing-characteristics of C2. Now, on port out1, instead of setting output-delays w.r.t C1 and C2, set them with C3 and C4. There would be no difference in the STA. However, this time, when we set the false-paths, do them with C3 and C4. This would cause the exact same thing as we wanted above, except that we cheated... and we did not cause other places where C1 and C2 are communicating to be treated as false-paths also.

Considering most designs these days are pad-limited (i.e. where chip area cannot be reduced because of the area taken up by pads), efforts are made by designers to reduce umbers of ports. This causes a junction of sorts, where different clock timing-paths intersect. This is a very good place to try out virtual-clocks.

Nice huh... thanks to my manager, now you know.

[Software] Source code management - golden rules?

2005-04-19T21:55:00.000+05:30

Frequently in any startup, there comes a time when the processes start mattering...or rather should matter more than the company. As Clayton Christensen says in his famous book

..at highly successful firms such as McKinsey and Company, the processes and values have become so powerful that it almost does'nt matter which people get assigned to which project teams.

I am at that career stage that I'm still not contaminated by manager-ese, which is to stay, vomit out improbable schedules. So here's my take on what should be followed during development.

Have a good spec

writing specs

Non Goals

wont

Have a good test-plan

I have no idea how to do either one. Anybody have any ideas?

Have good coding guidelines

writing code is easier than reading it

GNU coding guidelines

Use branching wisely

Eric Sink's model

No New Functionality on the Release Branch

Start breathing your code-freeze deadline atleast a week before date

All commits during these period should be associated with a bug-tracking number.

On release branch all check-ins must be reviewed

Release early and often

Ed Sim

Luca

[Business] Chinese noodles and Indian chutney

2005-04-11T21:11:00.000+05:30

It certainly is news when two of the world's fastest growing economies sit down together and sign agreements.
But what interests me most is the possibility of free trade agreements, especially those pertaining to cross-holding in the two countries. Such agreements would directly impact venture funding in Sino-Indian startups, because while several foreign investors want to invest in India or China, it is the slower growth in India or the unpredictability of governmental control in China, that dissuades them.
According to the Chinese Venture Capital Institute, venture funding in China is primarily by the government itself (48 % in 2003).
This may be good and bad. Governmental venture funding allows access to state-run projects and businesses, which may not be much of a deal in EU/US, but forms a large bulk of the market in China. It is bad, because bureaucracy as part of a startup is not that great an idea.
India, on the other hand does not have as large an appetite as China. While, this may reduce its attractiveness compared to our red-feathered neighbor, India is showing signs of a more stable growth.
Consider, the Incremental Capital-Output Ratio (ICOR). It denotes the percentage increase in investment (as a fraction of GDP) to produce a single point increase in growth. This means, having a lower ICOR is good for the economy. India's economy has shown signs of decreasing ICOR, from 4 to 3.6 (for the 10'th Five Year Plan). On the other hand, China's ICOR has gone up to 5.1 from 4.3 (between 1990-2003).
Both India and China have made huge investments in education, especially university level technical education. This provides for a large pool of sufficiently talented people to pick from.
Right now, we are running neck to neck in a race for technological supremacy (military and political are not my cup of noodles). But it would be far better if it were easier to run together. China has greater speed in governmental decisions. Its Special Economic Zones are great places to open new ventures.
I would say India has better technocrats - we can build and manage equally well (I like to think I'm one of them...only time and my boss will tell..very soon!!).
Beijing and Shanghai - two of the places I would definitely like to work. A few years back, I had a mild opportunity to work in Europe. It could have been great, but I could not afford to miss out on the Asian explosion thats about to happen.
Anyone interested?

[Personal] Carthage

2005-04-07T22:28:00.000+05:30

CIMG0195
Originally uploaded by sandys.

-> Hannibal was here <-

[Personal] Sun and sand...

2005-04-07T22:25:00.000+05:30

CIMG0220
Originally uploaded by sandys.

Tunisia at its quieter best...

[Business] VOIP and the Five 9s problem

2005-04-04T20:24:00.000+05:30

Ronald Gruia (and others for that matter) have written quite a bit about the VOIP showdown, which looks imminent. Off late Jeff has a focussed on the need for quality in VOIP service, something that is known in Dependable Computing as the Five 9s problem.
The question Ronald asked was a very relevant one - How reliable should VOIP be? Considering that mobile phones (in UK atleast) have an availability of 97-98%. In response Om Malik posted a with-all-due-respect post, with a figure: 100 %. There are just two problems with this - I think it might be unnecessary and secondly, it might prove to be impossible.
Let me analyse why it might be unnecessary. At this point of time atleast, VOIP is partly an early adopter product. But more importantly, it is a PIGS method. PIGS for all the non-desi's stands for Poor Indian Graduate Student, however it can just as easily be generalised to anyone else..iff they are limited in their communication expenditure but not in their need. It is enough for a person like that to be able to say "hi ma". In such cases, if I am the VOIP company, it becomes important to me to get to market as fast as possible and achieve sufficient penetration. Of course, granted that for most corporate use this might be somewhat of a problem. But then I dont see corporate users switching to VOIP en masse just yet. Consider this brilliant post by Tom Evslin about the problem of too much quality. About AT&T and mobile phones, he says

The six nines song deafened the company to the fact that consumers would trade quality for mobility

. Now, I would like to rephrase Evslin's law

The six nines problem may lead a company to alter its expectation of consumer expectation

. As Om Malik says "I have argued about the ease of use and reliablity before. VoIP continues to fail the mom test.". However, reliability and human-interface problems are orthogonal to each other. Consider what might happen if Apple decides to make iSpeak and jump into the VOIP arena - that would solve the "mom test". But would it need to solve the 5 nines problem?
Is doing this impossible? Bob Cringeley certainly seems to think so. Existing backbone providers might start traffic shaping to allow their own packets higher priority than, say a company like Vonage. But I'm happy if backbone providers give me 6 9s service. But as Bob, puts it in his later post, increasing demands on media content will result in the backbone networks being clogged. He suggests that media companies adopt a higher-latency download-and-play mechanism for media to free up bandwidth for low-latency VOIP, but let me pose a question - can Vonage wait that long? or can a student in California wait that long, to be able to afford calling Delhi?

[Personal] A Tale in the Desert

2005-03-30T14:41:00.000+05:30

I am speechless... in more ways than one.
Tunisia, land of ancient Carthage and Hannibal is an experience in itself. And so is the government, which somehow ends up with 99% of the vote. Therefore, my mouth was shut for the time I was in Tunis. I am speechless.
But, most importantly, what I learned in Tunis while working there, I could never have got by visiting or traveling in Africa.

People, especially young people, are the same everywhere. They like to drink, hook up with women and watch pirated movies. It was very much the same. The food is great there - with their couscous, lablabi and mlokhiya ( I hope I got that right). And I would be mortified if I did'nt mention harissa. Some of this stuff is unknown to the tourists who come here and stay for a few days, 'cos it is one of those "kaz" that is exclusively home-made. Oh and did I mention drink - Boukha, distilled out of figs and Sarab, distilled from dates. Yes, it is nice food-wise.

In their work, they are like Europeans, only much more rigid. I have been brought up in the Indian way of working, which is no doubt borrowed entirely from the American way. I am no stranger to arguments with my boss, late night pizza and all-nighters fighting for a deadline. Simultaneously, I have also enjoyed high pay raises and perks. Tunisian work culture is modeled after France's, with all its insistence on limited work hours, slow work and small raises. This is no small part due to the fact that the few companies to start up centres in Tunisia are French or Italian. I still remember the ritual that a person making eye-contact with you needs to come over and wish you and all the people around you. It is not what I am used to.

But they did have their quirks and idiosyncracies. Like the Arabic premium on virginity for the girl and not for the dude. And of course sentiments about the middle east run high...even in highly educated youngsters. That is somewhat scary, cos while in India and Pakistan, people in the street are only concerned about cricket and Bollywood movies, we dont hate each other. This is not true in the middle east. That is what scares me.

But what gives hope is the fact that Tunisia has one of the best education systems in the Arabic world. If the example of India is anything to go by, one can see wonders that a cheap and highly skilled workforce can do for the economy. But while cheap is very possible in Asia, it is very hard in Tunisia, where the people are paid according to Asian standards but pay for "kaz" (stuff!!) by European standards. Not to mention, the huge number of expatriate Tunisians working in highly skilled jobs in France.

To all my colleagues in Tunis- tisbaH 'ala kher

[Business] Long-Tail Software - the power of components

2005-03-21T14:28:00.000+05:30

Joe Kraus in his blog writes about business opportunities in the long-tail - the rarefied region of users who individually represent a small target, for a particular customisation of business process, but who collectively make up the bulk of the users.
Several companies have come up with several tools - purportedly to tailor to all these users. Joe Kraus himself plugs Jotspot. But in my opinion, the only type of companies who have actually been able to tap into this market are .. IBM, SAP, Inductis.
Oops... why is that?

In the business world, people dont look at a software - they way it looks, the way it can be customised, the way it can be changed - and then plan their requirements. It is usually..mostly, the other way around. We have learned this, the hard way.
However, it is possible to attain near 100 % customisability, with tools written in a way that allows them near independence.
For example, look at IBM. What are the tools that IBM uses for its business applications - Tomcat, Apache, php, perl. Each of this can exist individually. Each of this can be fixed individually.
What this means is a developer base, both within and outside the company that allows for almost complete compartmentalisation of performance issues, bugs, etc. Of course it is an easy argument, that components must be bound together very strongly to have predictable impact on performance, but this is very difficult to do in practice.
Look at Google for example. Its entire system is built for independent existence of different components. I do not believe that Google News is indexed in the same manner as Google search (given the Google API, why then havent we seen a better Google News by a garage hack).
When different components exist individually, can be run individually and therefore can be debugged individually, it is now the job of the sales guy to make a pitch to sell a solution, rather than a product.
Now it is much simpler - "I'll make you anything you want", rather than "You can do whatever you want with it".
I wish the best to Jotspot. Lets see if they give an API to program with.

[Personal] I'm officially official

2005-03-08T19:48:00.000+05:30

Hotblog has categorised me along with other guys like Steve Jurvetson, Jeff Nolan, Brad Feld, Ed Sim !
I mean, come on .. I was happy when Ed Sim posted a comment!!

P.S. Off to Tunisia for business. sun, sand and couscous

[Business] Exclusively underground

2005-03-08T19:20:00.000+05:30

A Bathing Ape is a Japanese fashion brand that is now slowly expanding its roots out of the Orient. Driven by Nigo, the 33 year old designer behind Bape (short for Bathing Ape), it is the hottest rage in Tokyo and whats more surprising is ...it has stayed there for some time.
All this is very interesting, except for one thing - how in the name of God, did someone make a cool ape?

Its similar to the story about Bose, I wrote sometime back. Its all in the magic of marketing.
Nigo used to sell clothes out of a sack at parties. It then became an underground fad. That one line makes all the difference - "It then became an underground fad". Do we honestly believe that an ape motif touched something so primal in us humans, that it became the thing to have? Nope.
Start with a clique - a group closely connected together. It is very, very important that the clique be cool people. If you are so unlucky to have your clothes worn by nerds ( I wear that badge proudly), you're in deep shit. Typical target cliques would be underground DJ's, street racers, hot women (not joking).
Once the product is introduced into the target clique, maintain exclusivity by making your stuff available only be referral to existing customers. This preserves the illusion of the cool people (closer to the center of the clique) getting cool stuff before other people, lower down on the fashion hierarchy.
Nigo and his Bape stores, still maintain that aura of exclusivity. His stores are unmarked and he does'nt advertise. I dont believe there is even a website of Bape.
This turns several management guru's theory of business on its head (and it does the same to my degrees-of-freedom theory). A long time ago, I wrote about friction-less business, based on a story by Steve Brotman. The point being businesses succeed when they remove friction out of their systems - this includes being more easily accessible to customers.
That does'nt work for "street cred". Google itself (the original inspiration for Steve Brotman's story) seems to have realised that. Their Gmail launch was so exclusive and so underground that people were selling free email invites for 50 $ on ebay.
Sony is now taking on Apple. It is NetworkWalkman vs iPod. Will they learn from their compatriot and preserve the veneer of exclusivity?
Of course, writing good code and having impeccable quality always helps..but that's another story.

[EDA] IBM goes statistical...Monte Carlo anyone?

2005-03-03T20:49:00.000+05:30

IBM has integrated statistical timing analyis on top of its EinsTimer static timing analysis tool. The new tool EinStat sits on top of its original tool and apparently gives a significant improvement in performance...

EinsStat analysis required about 18 seconds to optimize a test chip of about 3,000 gates, while a comparable Monte Carlo optimization took nearly 14 hours, IBM reported. Optimization of the 2.1 million-gate test ASIC took an hour and 10 minutes with EinsStat, IBM said.

Wow... that is really something. As an engineer working generally in the field of static timing analysis, I think the numbers are slightly off. Either EEDesign has a typo, or the IBM guys are reaaally bragging.
For example, consider this similar article from EEDesign: back in 2001, performance of the industry standard STA tool - PrimeTime took a couple of hours on a million gate design

n one Synopsys benchmark, a 1.3 million-gate design with 23,000 top-level coupled nets, and up to 2,500 aggressors per net, ran in five hours on a 400-MHz SparcOS5 workstation.

. Consider the slowest and most accurate of them all - SPICE. Benchmark results from a SPICE company shows benchmark results of approx 30000 gates at nearly an hour.
These benchmarks suggest that someone made a goof up somewhere. 3000 gates would not normally take 14 hours to time.

But even if it did good, thats good enough. Statistical timing analysis is seemingly hot in the world of EDA. However, its not just hot...its scalding.
Consider Monte-Carlo methods for optimisation. Monte Carlo methods model characteristics like circuit delay, capacitance and a whole load of other stuff inside simulators.
What happens is that since due to on-chip variations in fabrication, all the above parameters vary across the chip. If we want to model a chip accurately, we cannot take one value of the delay/capacitance/whatever but rather need a distribution function that tells us the probability of the delay/capacitance/whatever being a particular value at a particular co-ordinates (time, space, temperature). Then we need to evaluate this integral over 0 to n (where n may be size of chip/ time of operation/maximum operating temperature). It is in evaluating this integral that Monte Carlo methods are very helpful.
What they tell us is basically very simple: integrals are basically areas in a co-ordinate space. The integral of a particular function is the space covered under it. Now rather than evaluate the integral, we can arrive at an approximate answer using probability. Generate a sample co-ordinate and check to see if it is inside the area covered by a function.
Out of the total number of samples generated, the number of points which fall under the function will give the probability of the function.
This probability multiplied by the area of the bounded box around the entire function will approximate the area of the distribution.
Where is the trick - it is in generating a evenly spaced random sequence that will even out the points on the entire space. Such random numbers are called quasi-random numbers.
Statistical Timing Analysis (SSTA) tools ask the question differently. Instead of evaluating the probability that delay is less than a critical value in a design, they ask What is the maximum delay distribution from inputs to outputs. I do not know what is the computational order of complexity difference between the two methods, but apparently it is too good!

[Business] Enterprises: The Kingdom

2005-02-22T21:25:00.000+05:30

Chief Executive magazine has an article about Bose Corp.. It is the year of the celebrity CEO - first there was Steve Jobs (well there is always Steve Jobs), then there was Carly Fiorina (celeb status goes both ways).
Now it is Amar Bose.
Bose Corp. stands alone among other privately held companies. That is because, usually high tech companies need infusions of large capital - something that Bose Corp. has managed without. There are several large privately held conglomerates - Trump Real Estate, Virgin, etc. But all of these companies are in businesses which can do with a few hundred thousand of seed capital. As in, most of these companies are marketing based companies - you promise a buyer a few thousand feet of real estate, you go and find it cheap and you make a killing. But in high-tech you have an initial development time which guzzles money without a possible conclusion.
Q. How then do you do it?

A. By piggy-back riding on govt. money
Take Bose Corp. for instance - research that stemmed out of university research went on to become Bose Corp. Dr. Amar Bose's doctoral thesis was on complex variable theory, which was sidelined for the purpose of acoustic theory: a feat possible only by using the resources of the university.
Then there is the exclusivity aspect - not from the aspect of the serious audiophile. Nope, they would rather die than buy a Bose rig. It's custom, hand-built all the way for them. Exclusivity as an image projected for the housewife, the yuppie, the maitre'd. All of whom listen to music for say 2-3 hours a day. An illusion of exclusivity created by its resellers who are banned from ever comparing Bose with other competing products and who must provide Bose sound systems with their own exclusive space on the shop floor. Bose Systems are packaged brilliantly and though they may not actually work as well as the audiophile wants, it works and more importantly, looks good enough for me.
Why is this relevant to the original issue? Because Bose Corp. is again pulling the same rabbit out of its hat. Marketing a 24 year research in progress. Now who would'nt be interested in a research that spans 24 years. I would buy a car that has a suspension system backed by 24 years of research. Oh and just in case I worked for the competitor, I would be scared..very very scared.
Its really very simple - to create sex appeal for technology, it must appeal to women and early adopters. These are what are called the hubs of social networks. Appeal to them and you have planted the seeds that will sprout everywhere. Is it any surprise that no journalists have been allowed to sit in these cars. You have got to maintain the sex appeal till its too late for technology journals to pan them. Because you can always pay one of them to write good about you.
For a wannabe entrepreneur like me, it is an important lesson. I would like to be the carefree CEO who has no obligations to investors. But I do not think that is very easy. But it is entirely another matter to create an aura of charisma around your company and around yourself.

[Business] VC's have too much money ? Here's how to pitch

2005-02-11T22:36:00.000+05:30

Robert Cringley of PBS has this new piece about how VC's now have, suddenly too much money to give away. It is all because the funds which were invested in like crazy during the bubble are now closing. The managers either need to show investment or return the money with interest. Guess what they are choosing to do... Well it cant hurt, maybe I should start looking around!!

Somewhat related is this great article at UIWeb on how to pitch and idea. The best part I liked about the article was the elevator pitch.

... never allow yourself to believe your thing is so complicated and amazing that it’s impossible to explain in a sentence.

Some great examples were:

Inventing light bulbs: “It’s a way to make light from electricity.”
Defragmenting hard drives: "It makes computers run more efficiently"

another WoW (word of wisdom) was

Get out of your office / cubicle / apartment, and go find smart people you know to give you feedback. Ask them to pretend they are whoever it is you plan to pitch to

[dtrace] On demand debugging - Part 1

2005-02-11T20:54:00.000+05:30

One of the best things that I have seen the OS world is Solaris 10 and its associated debug tools. This includes - Dtrace, mdb, libumem . Together they can be combined into one of the most powerful debug frameworks that exists. Move over purify...
All that I have tried out is actually gleaned from user groups at various sun forums. But there is one thing Leventhal, Shapiro and Cantrill are gods.

Alright, dtrace commands are very similar to mdb commands. The syntax is
provider:module:function:name . The syntax is self explanatory, except the name part. This is a part which gives some part of the providers semantic meaning. e.g. exit will work when the function is about to exit. Some parts of the commands may be left blank.
for example the pid provider pid1000::myFunc:entry is invoked when the function is entered.

For me, the most interesting part is when, we need to debug loadable libraries. A little awkward, but very cool.

dtrace -qwc a.out -n 'pid$target:ld:dlopen:entry{printf("pid=%d loading: %s\n",pid,copyinstr(arg0))}'

This would print out the libraries that are being loaded. This small program hooks onto the pid of a.out using the pid provider. The $target is an implicit variable that gets the pid number of the a.out when it is run. Therefore consider the module ld. When the function dlopen in entered in this module (as part of the a.out process), dtrace will print the name of the library. Looking at the man page of dlopen, we can see that the first argument passed to the dlopen function (arg0) is the name of the loadable library. But we cannot print it directly. This is because dtrace operates in the solaris kernel space. To access names, etc. from the user-space, we need to copy them first to kernel space (via the copyinstr function).
When the library you are interested in has loaded, suspend the process. In another shell,

dtrace -n 'pid:library_name::entry{printf("function=%s probe= %s",probefunc,probename);}'

once dtrace has started its probes, resume the earlier suspended program.
The pid provider, will be the most used provider for any software developer interested in looking at the execution trace of his/her programs.

There is another feature of Solaris 10 which is of great help in locating leaks, debugging invalid memory accesses, etc. Maybe not so effective as valgrind or purify but hundred's of times as fast.
This is the libumem/mdb combo. Right now, I have only figured out how to detect memory leaks.

setenv LD_PRELOAD libumem.so.1
setenv UMEM_DEBUG default
setenv UMEM_LOGGING transaction

. This is necessary for the libumem memory debugging library to be loaded.

start the application using mdb


mdb ./a.out
>::sysbp _exit        /* to break before the program exits */ 

>::run
mdb: stop on entry to _exit
mdb: target stopped at:
libc.so.1`exit+0x14:    ta        8
mdb: You've got symbols!
mdb: You've got symbols!
Loading modules: [ ld.so.1 libumem.so.1 libc.so.1 ]

Here comes the interesting part. Finding leaks is as simple as :


>::findleaks
CACHE           LEAKED       BUFCTL          CALLER
someaddress      somenumber   impAddress     function address
----------------------------------------------------------------------
   Total       1 buffer, 24 bytes

>impAddress::bufctl_audit   /* to backtrace a particular leak  */
libumem.so.1`umem_cache_alloc+0x13c
         libumem.so.1`umem_alloc+0x60
         libumem.so.1`malloc+0x28
         main+4
         _start+0x108

>::walk umem_log|::bufctl_audit   /* to backtrace all leaks */

Whew.. end of day 1.

[Business] Mandates for Software Sales and Marketing

2005-02-06T15:41:00.000+05:30

Over the past few days, several posts in the blogosphere have mentioned, directly or indirectly, about marketing. These are posts which showcase and sometimes blur the distinction between marketing in software versus marketing toothpaste. For example, twin blondes in underwear would impact toothpaste sales more than software, which requires viral marketing like the Scoblephone. However, both toothpaste and software could do with some transparency in their manufacturing/coding process.
For example, Eric Sink talks about the tenets of transparency involved in being a Independent Software Vendor (ISV). Let me analyse some of his points :

Thou Shall Speak Out - I suppose this could be important in the viral marketing sense. I think this is going to work only now, when the power of the blogosphere is still in its nascent form. Once RSS/RDF/Blogs really explode, there will form a power clique or blog-o-cracy which will serve as a barrier for new blogs to break into the reader's RSS horizon. What will always work is injecting oneself into this clique - a process which will always expect that the member of the clique is involved in the development process. Human Psychology 101.
Thou Shall Let Others Speak Out - This is something very important. Users have to know that there are other users who are using this product. Have your R & D team reply to posts (Take a look at Bryan Cantrill of Dtrace) . But then why stop at Discussion Forums? Wiki anyone?? But for all this to be really effective, one has to take a long look at the next point
Thou Shalt Not Lie - You are not God. Dont hide your flaws. Give the impression that you are listening to your users. Get your R&D to reply.
Thou Shall Make Thy Customers Lives Easier - It is important that the path to adoption be as smooth as possible, from medium of distribution (Bittorrent?) to demos, support, bug-tracking - especially your licensing schemes. It is relevan here to point to a post by Frozen North about usability and marketing. Take the IPod Shuffle for instance. Its competitors have blasted for having a limited feature set. Falling into the feature-set trap is easy. For me it is best illustrated by the GIMP image program. Its is overflowing with features, however it has a lot left to be desired from the usability standpoint. For Apple, usability is marketing. Throw away the features that do not flow together. Less is more. To give a top-notch user experience rather than feature-list, the Shuffle threw away compatibility with anything other than iTunes.
Thou Shall Tell Thy Company's Story - The old management adage : underpromise and overdeliver. Very cliched, but works. Customers want to know about who you are and can they trust you to be around 10 years from now.

Selling software is increasingly becoming all about openness, because unlike toothpaste, software is not tangible (wow Morpheus!!). Selling something intangible takes more than a spreadsheet and a two-color ad.
Ed Sim talks about a "repeatable sales model", in one of his recent posts. In reply to my question about, what seemed to me, a management buzzword, he says

...building a product that solves the problem for one, two, or three customers is different from building a product for many customers in a market that is large enought to support your business and others who are already there or will show up.
Think about whether or not the product as it exists today solves the problem for a number of different players or did if it required significant customization to the extent that the product needs and requirements were different for each customer.

The "repeatable" part means that you have nailed down the value proposition, know who in the organization you are selling to, can cost effectively reach that person, understand how to win deals against competing products, and can use this as a model or template for additonal sales headcount.

Another point made by Ed, can be fitted on to the software marketing setup - release early and often. This is the cathedral vs the bazaar model, popularised by the linux kernel release model. In conjunction with the point on (Thou Shall Let Others Speak Out), this would be a marketing mechanism directly interfacing with the R & D team.

This is what I am used to doing.

Frequently it is OK, customers are people and are usually civil. It is just that one big customer who has the power to raise a big stink, be rude and in general make a nuisance of themself. That is when everything begins going downhill. It is then you realise what it takes to keep the customer on his throne.

[Startup] Iterative Entrepreneurship

2005-02-01T11:56:00.000+05:30

While browsing through the VentureWiki archives, I came across this article by Ross Mayfield about successful startups. The article sites a study by DarwinMag, which in turns refers to a report by Crescendo Ventures.
Notably, the one thing, they all emphasize is iterative business - in other words, tweaking the business as you go along...something I could'nt understand. What to tweak and how ???
Actually the Crescendo Ventures report was quite interesting. The report refers to a study on storage device manufacturers (read hard drives), in particular 456 storage vendors formed over 30 years. It was interesting since the Innovator's Dilemma focuses on the very same thing. As Clayton Christensen notes, storage devices are the fruitflies of the technology world - their short lifecycles make them ideal study subjects.
However, while Christensen focuses on the disruptiveness of innovation, this study focuses on the fact that most storage ventures are not disruptive at all. This is interesting to me, with all due respect to Mr. Christensen, because while this may not bode well for a long-term sustainable enterprise, it does offer value to the entrepreneur in terms of a short term exit strategy. Table 20 of the Crescendo study does list several of these companies and rather than appeal to several customers, I fail to see how they could have disrupted the storage business itself.
But then, what about iteration? As far as I can make out, it means to continually showing potential customers your solution and iterating towards the golden solution. As I read somewhere, this may be risky (as in loss of the cool idea), but it is an acceptable risk. Customer is God. Listen carefully.
This also means, that tracking risks carefully and closely. For example, on launch of a bluesky project in a company - the expenditure, behavior and trends of target market, customers, etc. should be tracked very carefully. Weekly business estimates should be recalculated and matched with expenditure to re-evaluate viability.
Two other interesting things that I learned from the report were that
early stage startups need not be handicapped for lack of a complete executive team. Focus should be on technical prowess rather than marketing wizardry. Secondly, involvement of a proven business development executive is a necessary ingredient for success. This is interesting, for VC's are likely to look for this.

[Business] The New Art of Bootstrapping

2005-01-29T11:54:00.000+05:30

This great post by Tim Oren, about the new generation of software startups, made me sit up and start thinking...again.

Now we've taken a step back closer to those early days, as the combinations of open source and standards, and commoditization of hardware and software, once again make a low rent market entry feasible.

How true. Even in the mathematical and formulaic world of design automation, people prefer to work via open standards and tools (GCC, Flexx, etc. ). Which is why a move by Cadence to release an OpenSource circuit database library (OpenAccess) is a very cool thing.

I'll try to adapt Tim's opinions and views to the EDA industry. For a new company to break into the EDA industry, it has access to open standards like the SDC (Synopsys Design Constraints), SDF (Standard Delay Format) and now - OpenAccess. Therefore, it becomes quite possible (not easy) for a new company to build tools for validation, cleanup, verification, housekeeping, etc. on the basis of these standards.
However, while on the one hand, building a product first and then searching for customers is quite an easy proposition for those in the web-app domain, I do not believe that is quite easy in the chip-design domain. The trend has always been to have contacts first and products later. For example, while a two grad-student team can beat out a google, by sitting in their garages, I do not believe Intel will even take a second look at me, if I do that. I think that's because EDA customers are concentrated in a particular cities, companies...even buildings. The target population is not a anywhere-anytime kind. Primarily this is because of the need for powerful hardware for processing tasks (read server farms). What I cant still understand is - Can I find a way to offload processing tasks from the CPU to the GPU? The GPU is a vector processing architecture, however there could be a way.
The second important task would be to have a way of making access to my technology, a path of least effort.
Is web enabling the answer, I do not believe so - primarily because of the horrendous computation power required. What is the magic bullet?
The second alternative is to already have a reputation in the industry. Not unlike every other industry there is. Look at Sierra Design Automation, Magma, etc. They were founded, and hence gathered credence, from the prowess (real or imagined) of their founders.
In any case, I think Tim hits it on the mark when he says

Build as little as possible, as fast and cheaply as possible, while demonstrating some unique value.

This would need to be done in association with a pre-customer. Else, such fine-grained precision to hit the requirements on the dot, is simply not possible.
And once I have VC investments - well that's another story, I dont know anything about...yet.

[Business] Mousedriver chronicles

2005-01-26T20:27:00.000+05:30

I swear I did'nt just make that name up. It is a name well known, apparently, to entrepreneurs and VC insiders alike. The Mousedriver Chronicles are a series of newsletters written by two just-out-of-Wharton graduates - Kyle Harrison and John Lusk. It was supposed to be a honest and humorous take on their life as first time entrepreneurs making... computer mouses shaped as golf drivers.
These were the heady days of the dot com boom, when every fresh faced grad with an MBA was opening a portal (I would know some of them too). But these two grads built a company making a cool product, but without any dot-boom sexiness.
The Mousedriver Chronicles is now available as a book, and the much hallowed Brad Feld has it on his read-recently list.
All in all, a good read.

[Business] Incidental Prices, Software and Degrees-of-Freedom

2005-01-22T23:36:00.000+05:30

There is an interesting story in the Journal of Marketing Research. It talks about the effect of incidental prices on purchasing habits. But are these relevant for software business ?
The paper talks about a study about the effects of incidental prices - price of a shirt - on a customer's purchasing decision - purchase of a CD. Both products have nothing in common, but they still affect purchasing decisions.
I was reminded of this feature by Joel Spolsky, software guru extraordinaire. I have written another article,which uses the same source, however in a slightly different context.

As Joel says, software purchases are very dependent on their complements. I like to think of complements as...degrees of freedom. Given a certain product, what are its degrees of freedom. For example, a portable mp3 player can be used to store mp3s and store Powerpoint presentations. So, it has two degrees of freedom. Imagine that I write a piece of software which compresses and encrypts data as it is being downloaded onto my iPod. That introduces a third degree of freedom.
People buy software, when they can increase their degrees-of-freedom at lower prices. How is this relevant? I'd like to take it further by suggesting that people buy software when they can perceive an increase in degrees-of-freedom at lower prices.
So, I may not really have a need for encrypted mp3s... I may also never buy this software. But because I know, that I have an extra degree-of-freedom available to me , that knowledge has an impact on my purchasing decisions - I may actually buy something more expensive, because something else was priced lower. Merchant pricesmay be adjusted according to this.
While we are on the topic, we could also analyse another aspect. This is relevant in the context of marketing because the paper suggests that this could be a good way of targeting online customers.

While opening a browser with the intention of buying a book at Amazon.com, a pop-up advertisement that touts flights at Orbitz.com “starting at $124,” could make shoppers less price sensitive.

This trend, if indeed used, could very well backfire. Because, directing airline ads at a book shopper are indeed a poor way of targeting. Moreover, it is not fair to the airline. A much better option would be to identify the book's complements. But, in my limited experience, a textbook can also have degrees-of-freedom. For example, if you finish the First book of Calculus, you may understand the Second Book of Calculus.And , if you buy them both together, not only do you get two degrees-of-freedom, you get a discount!!
P.S. A good article on pricing strategies primer is a feature by Heidi Cohen.

[Software] Thread and register level debug using Solaris adb

2005-01-21T20:50:00.001+05:30

gdb is a good tool..usually that it. But my experience has shown that gdb has issues when the problem is in symbol resolution, traps, etc. which occur at load time.

For example, given an application that actually loads a shared library at run time. If for some reason, there is an issue in symbol loading itself, gdb will not be able to instrument the code properly. Such an issue is not very common, but it occurs in certain cases. For example when you link a static library into a loadable library, there might be issues.

Then there is adb itself. It is not as easy and well documented as gdb. First of all, commands are not stuff like "run", "debug", "break", etc. For example, the command to quit an adb session is $q . Secondly, it has no prompt by default. (You could give it one using the $P command. )

Basics

To start adb, you use

adb [binary file]

To run the binary, you use the command

:r [command line args]

On crash, the backtrace comes up using $c

Breakpoints

Breakpoints can be set using

function-name + offset :b

The offset is the number of bytes after start of function.
On hitting a breakpoint, the program may be stepped using :s or ,10:s(to step 10 instructions). Program execution may continue to next breakpoint by :c.
At any breakpoint, the program execution trace may be examined by $c. When doing so, often we can see function calls and the arguments passed to them in hex. To view the contents of the address pointed to, by the hex, use (hex value)?s (for address) and (hex value)/s for data. the s modifier at the end denotes values should be shown as string. You could as well use d (decimal), x (hex), etc.

Examining data

To examine the next 10 instructions from current location (breakpoint or abnormal termination point) is <pc,10?ia. The < modifier actually denotes that you are operating on individual registers (in this case the program counter - pc). Now, for example, if you do <o3,10?ia, adb will pick up the value in register o3, look at that address in memory and check if there are program instructions at that memory location. so we are looking at the contents of memory pointed to by o3. To look at the contents of o3, you need to use <o3=x. This will print the value contained in o3.
To print the next 10 addresses , starting from address of function f1 is to use f1,10/ia.

Usually, during debugging, its sometimes good to remember the following info about typical registers:
%l0 %l1 %l2 %l3 %l4 %l5 %l6 %l7 (not affected by function calls)
%o0 %o1 %o2 %o3 %o4 %o5 (used for passing arguments into functions, usually by moving something in them)
%i0 %i1 %i2 %i3 %i4 %i5 (used within a function to access arguments and return values)
%g0 %g1 %g2 %g3 %g4 %g5 %g6 %g7 (globally accessible registers)

At any time contents of some typical registers can be seen using $r or $?

[Personal, Business] Powerpoint and the art of VC presentations

2005-01-18T11:49:00.000+05:30

Office clipart
stock.xchng - for lots of free slide images (wow!!)
Beyond Bullets - for all your un-bulletizing

Brad Feld's article(reproduced here for fear of losing it!!) (also by Seth Levine ??)

1) WHAT IS YOUR VISION?
- What is your big vision?
- What problem are you solving and for whom?
- Where do you want to be in the future?

2) WHAT IS YOUR MARKET OPPORTUNITY AND HOW BIG IS IT?
- How big is the market opportunity you are pursuing and how fast is it growing?
- How established (or nascent) is the market?
- Do you have a credible claim on being one of the top two or three players in the market?

3) DESCRIBE YOUR PRODUCT/SERVICE
- What is your product/service?
- How does it solve your customer’s problem?
- What is unique about your product/service?

4) WHO IS YOUR CUSTOMER?
- Who are your existing customers?
- Who is your target customer?
- What defines an "ideal" customer prospect?
- Who actually writes you the check?
- Use specific customer examples where possible.

5) WHAT IS YOUR VALUE PROPOSITION?
- What is your value proposition to the customer?
- What kind of ROI can your customer expect by using buying your product/service?
- What pain are you eliminating?
- Are you selling vitamins, aspirin or antibiotics? (I.e. a luxury, a nice-to-have, or a need-to-have)

6) HOW ARE YOU SELLING?
- What does the sales process look like and how long is the sales cycle?
- How will you reach the target customer? What does it cost to "acquire" a customer?
- What is your sales, marketing and distribution strategy?
- What is the current sales pipeline?

7) HOW DO YOU ACQUIRE CUSTOMERS?
- What is your cost to acquire a customer?
- How will this acquisition cost change over time and why?
- What is the lifetime value of a customer?

8) WHO IS YOUR MANAGEMENT TEAM?
- Who is the management team?
- What is their experience?
- What pieces are missing and what is the plan for filling them?

9) WHAT IS YOUR REVENUE MODEL?
- How do you make money?
- What is your revenue model?
- What is required to become profitable?

10) WHAT STAGE OF DEVELOPMENT ARE YOU AT?
- What is your stage of development? Technology/product? Team? Financial metrics/revenue?
- What has been the progress to date (make reality and future clear)?
- What are your future milestones?

11) WHAT ARE YOUR PLANS FOR FUND RAISING?
- What funds have already been raised?
- How much money are you raising and at what valuation?
- How will the money be spent?
- How long will it last and where will the company "be" on its milestones progress at that time?
- How much additional funding do you anticipate raising & when?

12) WHO IS YOUR COMPETITION?
- Who is your existing & likely competition?
- Who is adjacent to you (in the market) that could enter your market (and compete) or could be a co-opted partner?
- What are their strengths/weaknesses?
- Why are you different?

13) WHAT PARTNERSHIPS DO YOU HAVE?
- Who are your key distribution and technology partners (current & future)?
- How dependent are you on these partners?

14) HOW DO YOU FIT WITH THE PROSPECTIVE INVESTOR?
- How does this fit w/ the investor’s portfolio and expertise?
- What synergies, competition exist with the investor’s existing portfolio?

15) OTHER
- What assumptions are key to the success of the business?
- What "gotchas" could change the business overnight? New technologies, new market entrants, change in standards or regulations?
- What are your company’s weak links?

NWVenture Voice's 5 golden tips (again reproduced here.... apologies..but would hate to lose it)

1. Outline
First, have an outline. Be organized. The best top level outline I have heard it from one of the super masters of presentations, Jerry Weissman. Before you focus on all the snazzy charts, make sure you do the following:
• Tell them what you are going to tell them: Show them where you are going to take them, on the title slide.
• Tell them how you are going to tell them: Have an agenda slide and stick to it.
• Tell them: make sure the body of your presentation always reinforces your opening point.
• Tell them what you told them: wrap up, recap and go for the close.

2. In a nutshell
One great tool for making this organzation stick is what I call the "in a nutshell" slide. This is using your agenda slide to tell the skeleton of your whole arguement. When presenting to Steve Ballmer it often happened that you never got off the first slide after the title, so make sure it really works for you.

Normally, I like to see In A Nutshell slides that act as a template. On one side they highlight, even number the key elements of your story/pitch/arguement and in parallel on the other side they give the top support points in summary. As you then move through the deck you keep the left hand template to reinforce the whole arguement and help people remember where you are in it.

3. Clear, simple case
Show why your company/investment should exist in the first place. Do the simple case using what we call your ABCs or situation/gap analysis. Where:
• A = Today: the current situation in the market/big growing
• B = Tommorrow: the place the market should be/juicy opportunity
• C = Gap: what's missing to get to B/the special play you are poised to make to fill it and win

4. Simple positioning and proposal
Then tell why your way of filling this gap is better than everyone else's. One simple outline for this is what we call the XYZs - "We are the only X company/product that solves Y customer problem in Z unique way," where
• X = your category: critical for VCs, we need to put you in some box, to make comparisons; never invent a category, improve one.
• Y = the target: the buyer, the person who actually writes the check, great if you actually have some.
• Z = your differentiation: your advantage, or the key positive distinction you have over your competition.

It also helps if you can back all this up with real support, like your team's track record, customer traction, a real competitive analysis (thier ABCs), etc. A demo is not enough. Proof is better than claims.

5. Best foot forward first and strongest
Tune the organization of your story to the stage of your company. And always put the strongest stuff upfront.
• An EIR: It's all about YOU and the market opportunity/competitive gap.
• A seed: It's all about initial market validation (quotes from friends with important job titles in your target customer's industry), then about the product spec, the team and the above.
• A round: It's all about initial customer traction and economics - some demonstrated willingness to try and pay - show the best real numbers you have, then about the product itself relative to others, then the above/
• B round: It's about momentum - show the sales numbers, the trends and the economics, then all the above.

Then of course have a well thought out and agressive enough ask.

[Personal] Media Lab Europe to close down...

2005-01-17T23:07:00.000+05:30

One of the places, I so wanted to work at. Great and eclectic bunch of people. Which is always a good thing. Except for one thing.

Leaders and Vision

Media Lab Europe R.I.P

[Business, Asia] CIA's strategic 15-year vision document

2005-01-15T20:29:00.000+05:30

The National Intelligence Council's document on Project 2020 highlights the possible growth of Asia, and especially India, in the next 15 years, in all fields including business.

As expected, China will become the second biggest defence spender after the USA. However, the most interesting aspect was the future of global and asian business. China will continue to be the engine for world business growth. However the existence of strong domestic capital markets in India coupled with a strong knowledge based economy will cause India to retain stability in growth.

To justify my assertions, I decided to take a look at Manufacturing News. An interesting news was the relocation of chip manufacturing fabs from Taiwan to China. Since, I have an insiders knowledge of that particular industry, I know that Taiwan is famous for its reliant chip fabs. TSMC is well known for its Reference Flows which validate approved design flows for tape-out. Which makes this move all the more surprising. Why?

Another news source I found, refers to a proposed 2.4 billion USD investment by the Chinese Govt. in power. Fabs are ultraclean environments, and extremely power hungry. Enchancements to infrastructure by the Chinese govt. come as a lifesend to the chip manufacturing industry.

Simultaneously however, a parallel news item listed the growth of the Bangalore Intel design centre as the largest design centre outside the US. What does this imply in terms of economic stability?

A chip manufacturing plant is much more human-resources intensive than a design centre. Any fluctuations in global technology markets will be much more pronounced in China, than India. According to Goldman-Sachs, India-China business is projected to rise fast from the 7.6 billion USD, it is now. This growth has to be in conformance with the synergy of knowledge and manufacturing industries working together.

One of my earlier posts makes a case for startups in India. What worries me about China is the manufacturing base. In a good year, their capital returns would dwarf India's. However, this capital may be used to power their own knowledge economy. There's only one thing we can do - never stop running.

[Business, Software] Developers, Open Source and Complements

2005-01-10T22:22:00.000+05:30

Ed Sim of Dawntreader has a post about roping in developers as a "buzz"-generating mechanism for new projects. He says

While it is imporant to court developers and technologists in the sales process since they typically have to give the technical buy-off and can just as easily squash an opportunity, it is not a great and economical use of time to have your most expensive direct sales resources and sales engineers doing this.
Enter the web and the open source movement.

Interesting. However, going through his post reminded me of another post by software guru Joel Spolsky. In his article, Joel talks about substitutes and complements. For example, consider a graphics card by nVidia. The sales of this card will go up, when its complements (read computer games) become cheaper. What's the point, you ask...

Using the web as a platform for building community is well and good. But, what are the complements of a brand new product - nothing. For example, I could develop a device which connects my iPod to record my radio shows. Will it sell? I dont know..even though in itself it is a complement to iPod, I still have to create its own set of complements.

Can I depend on the developer community to do it for me. Nope. There has to be a momentum, which has to be created by the creators. So for example, if I had used a hackable linux kernel in my device - I could write modules to compress the stream on the fly/encrypt it/broadcast it to multiple receivers. Can I now depend on the deveopers? That depends on how much you maintain the momentum.

One of the best lessons I learnt from my Big Boss was "respond fast and release often".

Which brings us to a quote by Joel:

..when an economist considers price, they consider the total price, including some intangible things like the time it takes to set up, reeducate everyone, and convert existing processes.

Creating a web based community is well and good. But a community withers out fast if there is no training/support. Companies may think that they have a cheap medium of the internet for marketing/sales. But customers now expect faster and faster responsiveness to issues. That does add on to cost.

The web can never equal an attentive salesman who can sense body language. Its what I have learnt from deploying applications at customer sites.

[Software] Functors and expression templates

2005-01-08T19:31:00.000+05:30

Consider a scenario in which, given a graph, you have to traverse it. How do you do it - write a function. Now consider a situation, in which only those nodes of a graph have to be touched which have a weight of say 8. Again, very simple - just pass an argument to the function (of value 8).
But what if, the function is actually a callback. That is, the function is passed to another function. In such cases, it becomes difficult in terms of maintainability and performance (more on that later). Now, if you could somehow create a function with state, the above becomes so much easier. Enter functors....
Functors are also frequently better performing than functions.

Stroustrup says:
"A suitably-defined object serves as well as - and often better than - a function. For example, it is easier to inline the application operator of a class than to inline a function passed as a pointer to a function.
Consequently, function objects often execute faster than do ordinary functions." Stroustrup, TC++PL (3d. ed.), Sec. 18.4, p. 515.

Functors can also be used in recursion. It took me a while to figure out the correct syntax to do this:



class sssFactorial{
    int operator() (int val){
         if(val==1) return val;
          else return (val*this->operator()(val-1));
     }
};

This is well and good, but as we can see it is not exactly very trivial to construct functors as compared to plain vanilla functions (writing those extra lines will put a strain on your poor fingers).
There are two ways of handling this - generating functors by using:

Functor binding
Expression templates

Functor binding essentially binds a binary functor (which takes 2 arguments) into a unary functor (which takes one argument). How is this good you ask - how can reducing the degrees of freedom be any good, right?
Wrong.
Binding may reduce degrees of freedom, but it makes a new functor from an old one.
Suppose you were given two functors prod (to find the product of two parameters), div (to find quotient of two parameters)- how would you make a functor that creates (x*b)/c
f=bind1st(bind1st(div(),b),c) ;
Now, f(x)=(x*b)/c

(The C++ Standard Library also has the compose functor which is powerful still)
Expression templates is a C++ technique for passing expressions as function arguments. A compliant compiler would be able to inline all the expressions (without the need for temporary variables), thereby speeding up code.

Expression templates are one of the most powerful features available in C++. Consider a matrix or vector operation of the form

X=Z*Y + Y,
Where X,Y ans Z are matrices.

In this case, at each step of the computation of the mathematical expression, temporaries are created, each of which is of matrix size, which hogs the available memory. However, it is possible to decompose this complex operation into mathematical expression involving primitive ints/floats (i.e. actually determining the matrix formula for calculating each element of the resultant matrix). This is good, very good...but..not very maintainable (the horror..the horror)
Enter expression templates.
Expression templates allows the compiler to generate a compile time tree of expressions. No more trying to write large expressions, the compiler does it for you.
Along with inlining most of the functions/functors used - producing faster code.
For a valid, compilable code - see below...way below.
For example several high performance libraries for matrix operations use expression templates (for example, MET ).
Expression Templates are a great way to calculate matrices, with great performance.

In EDA, matrixes are used for simulation, for functional verification, petri nets and hundreds of other applications. Cool..or what!!

Code for a sample expression template - just gives value of x*4



#include <iostream> 

#include <stdio.h> 
using namespace std;

     
template<class A>
class Expr{

public:
     Expr(const A& a):_a(a){}     
     //needed for variable
     Expr(){}       
     int operator() (int x){
           return _a(x);
      }    
    int getVal(){return _a;}

private :
      A _a;
};
     
class Variable{
     //no constructor needed, since this is a variable
      public: int operator() (int x){ return x;}
};

     

class Constant{
public:
      Constant(int a):_con(a){}
      int operator() (int a){return _con;}

private:
      int _con;
};
     
template <class A, class B, class Oper>

class BinaryExpression{
public:
      BinaryExpression(const A& a, const B&b): _a(a),_b(b){}
      // The functor calls a static inline function

      int operator() (int x) {
           return Oper::apply(_a(x), _b(x)) ;
      //remember _a and _b need not be constants, they may be expressions.
      }
private:

      A _a;
      B _b;
};

class Multiplier{
      public:Multiplier(){}
      static inline int apply(int a, int b){ 
       return a*b;
      }
};
     

template <class A>
Expr <BinaryExpression<Expr<A>, Constant, Multiplier> >
operator* (const Expr <A> & e, int val){
      //The most important line. This creates BinaryExpression object, 

      //wraps it in a Expr and returns it.
      //While doing so, it wraps the constant operand in a "Constant" object.
      //The Multiplier functor is passed to BinaryExpression for evaluation.
      return Expr <BinaryExpression< Expr<A>, Constant, Multiplier> > 
           ( BinaryExpression<Expr<A>, Constant, Multiplier > (e,Constant(val)) );
}


//the evaluation function.
template <class anything>
int evaluate(Expr<anything> e, int val){
        cout<<e(val)<<endl;
}


int main(){
     //We create a variable and use it.
      Expr <Variable> x;
      evaluate(x*4, 5);
}

[Business] Stanford Edcorner

2005-01-08T01:25:00.000+05:30

Found this great site on stanford :

http://edcorner.stanford.edu.

21 lectures by Vinod Khosla!!! wow

[Business] Financing for startups

2005-01-07T23:33:00.000+05:30

Brad Feld has a great post on venture financing and term sheets for startup funding. He calls it ne of his super-top-secret VC magic tricks .
Brad writes about pre- and post-funding valuations of startups. An important point he made was that increase of a stock pool would reduce the valuation. A stock pool is increased at valuation time to make accomadation for employee compensation. I can understand how this can dilute the investors share of the pool, but I dont understand why the overall pool must drop. Here's an excerpt from his post

...note that the size of the pool is taken into account in the valuation of the company, thereby effectively lowering the true pre-money valuation.

Another piece of advice that he wrote about was about steering clear of warrants in first round of funding. Which made me really curious as to why. On further study of other sources, I encountered other jargon - bridge loans . Putting it all together, I understood that all these are becoming more commonplace these days with VC's demanding "staged investment". I can personally identify with these VC's since technologies like mine (EDA) are very niche and customer driven. Without proof of a concept on core technology, or milestones in terms of actual deployement in companies, it is very difficult to part a VC and his money. However, there is a variation to this concept, wherein the VC invests the full capital in a given round, but may require anti-dilution warrants that are triggered if certain targets are not met by the management.
In certain cases, the VC's may extend loans, as a stop-gap measure in between series of funding. For these types of investments, warrants are issued which give the lender the option to make an additional investment (at terms which are established in the next round). This is beyond the fact that the principal and accrued interest are converted to stock in the next round.
Sweet.
I suppose that complexity is the real hurdle here. But at the risk of sounding juvenile, is accepting a lower valuation in exchange for no warrants a good thing. Previous valuation act as a benchmark for the next rounds of valuation. I would rather say that warrants would be necessary if both you and your VC's want a closely knit family. Personally, I would rather issue warrants to VC's who get what I am doing, rather than worrying all the time about valuation. I believe there is plenty of money for everyone.
I think Feld hits the nail on its head in his last paragraph - choose your early investors wisely, they can hurt next stages of funding.

[Timinganalysis] Negative setup and hold

2004-12-31T22:08:00.000+05:30

A negative setup and hold condition is a very confusing proposition in static timing analysis. Support for this type of conditions was added in the Verilog LRM, only in the late 90's (using the $SETUP and $HOLD constructs).
The basic idea is something like this:
Consider a module with an ideal flop in it. Now, there exists a data path (from primary inputs of module to D of flop) and a clock path (from primary inputs to CLK of flop). Suppose the data path delay is DD and clock path delay is 0 . Therefore, if we consider the clock pulse reaching at the primary input of the module as the reference time, the clock pulse will reach CLK pin (of flop) at 0. The data pulse will reach D pin at DD. Therefore, for setup check to be met, the data pulse must reach the primary inputs of the module, at -D, which means the setup requirement is D.
Now consider a clock path delay of CD. This means that the clock pulse now reaches the flop, only after time CD. This means, the data pulse need not begin so early, and rather it has to begin at -DD+CD time(just right shifting the pulse by CD time). This means the setup requirement is now DD-CD. In this case, if CD>DD, then the setup requirement becomes negative, which means, the data pulse can reach the primary input of the module after the clock pulse has reached there.

Similarly for hold: Consider that the data delay is 0 and the clock delay is CD. Now, the data must not change for atleast CD time, for the flop to be able to latch it. Therefore, the hold requirement is CD. Now, consider a data delay of DD. This means that, now the data need not change only for CD -DD. This is the new hold requirement. If DD>CD, then hold requirement is negative.

If we analyse these results mathematically, we can see that setup relation + hold relation =0. Physically, this implies that an infinitesimally short pulse (a delta pulse) can be captured; which is of course not possible. A more accurate model would be:setup_val<DD-CD (for setup to be met, the time at which data begins should be atleast DD-CD before 0)
hold_val < CD-DD (for hold to be met, the time for which the data should be stable should always be greater than the hold_val)

Now, the model we described, regarding the module with an ideal flop, is actually a real world flop. In an actual flop, there are more than one data paths and 8 clock paths. Therefore the more accurate description would be: DDmax-CDmin >= setup_val (for setup to be met)
CDmax-DDmin >= hold_val (for hold to be met)

These kind of relationships, especially the ones, where a negative relations can hold cause problems in simulators. Take for example a data pulse, which rises at 0.0 and falls at 2.0. Now the clock pulse rises at 3.0 . Lets say data delay is 1.0
Assume the origin at the clock pulse (3.0) . Therefore data rise is at -3.0, fall is at -2.0 .
The setup relationship may be specified as 2.0, which means data should be present at 0.0-2.0=-2.0 . Now data will arrive at -3.0+DD-CD=-3.0+1.0+0.0=-2.0 (setup OK)
The hold relationship may be specified as -1.0, which means data must not change till 0.0+(-1.0)=-1.0. Now, according to our relationship, data will not change till 0.0+CD-DD=0.0-1.0=-1.0.
All looks hunky dory...but...
There is no problem with the timing checks, however in software, the simulator would capture the falling 2.0 edge rather than the high edge. So the simulator will get the functionally incorrect results, though timing accurate. If both setup and hold relationships were positive, then this would never have happened. So now what ?
Very simple actually, instead of taking an ideal clock, the simulator takes a delayed clock. Therefore all calculations are done wrt this delayed clock (in the above example clock is delayed -1 wr.t data), so the simulator will not latch the falling edge.
whew...

[Business] (dark?) Angels of Business

2004-12-30T21:02:00.000+05:30

To all the victims of the tsunami.. may your souls be at rest.

Inc.com has a story about angel investing, and how this landscape has changed. In an earlier post, I wrote about VC's and how their expectation horizon has changed (maybe I was off the mark a bit - but Ed Sim actually commented back!! Thanks Ed..). The fact remains that VC's and angels alike are pushing for faster exits. The company in which I work currently (in the EDA industry), is a small company of 100 employees. We are still taking investments from VC's, but I sense a vibe of edginess as people anticipate sell-off conspiracies.

For example, the Inc.com story talks about a second time entrepreneur who is being pressured to exit in 5 years. So what does it say for a guy who wants to build a sustainable enterprise? In my country, all the biggest industrial houses (with revenues of almost a billion) - Infosys, Reliance, Bharti - were built from scratch. However, they were not sold off to, say SAP, Exxon-Mobil, Vodafone respectively. I guess its because of the inertia in our culture, and not in a bad way.

I think that is one of the biggest deterrent that international investors find in India. There is a lesser chance of scoring a hit-and-run here than it is in, say China. I belong to the generation that has seen India transition from a mostly hand-to-mouth economy to the world's fourth largest. I believe this has taught me, atleast, to focus on sustainable enterprise, rather than make a fast buck. Bu most importantly, my field does not let me have a "positive cash flow in 12-18 months".

Does that mean that I have no chance of getting angel investors? I think so. which is why it becomes important to find rich and smart friends. ( note to self: Ph.D. or MBA ???)
I think industries which are more rigorous tech rather than nice customer-centric toys (forgive me flickr), will just see a handful of angels (hence the rich friends)

[Personal,Software,Design] Bell Labs India

2004-12-23T23:53:00.000+05:30

Red Herring reports that Bell Labs has opened its latest research centre in Bangalore. I have been a personal fan of Bell Lab projects, especially the work they do in topology. I know vaguely that a lot of the algorithms of discovering nodes in a network were developed at Bell.

One of their sample projects - Mediation adaptor generation tools - looks interesting.
I wonder if they are using any shell-based interface; most projects which do automatic generation of code, rtl, specification, etc. fail in the hands of the user because they are too cryptic to use.
Suppose, this input specification had to be given in UML format and then the software runs and spews out some cryptic output. Of what user is it to me as a user none.

The basic problem with most software is that it either gives too much information all at once and that user cannot do anything interactively. These are the two things that I learnt working with my present employer.
Workarounds:

Too much info: The tool must give messages in a kind of keyword-mapped message. For example , instead of saying "Node 123 mapping complete", it must say "[NodeProtocol-23] Node 123 mapping complete".This way, the user has to just remember that he saw a keyword : NodeProtocol-23: and can look it up, grep for it, save it in a named file.
Too long info: The screen message should only be 2 lines long. The rest of it should be looked up in a long-help documentation. The user is smarter than you think
Interactivity: suppose, I remove Node 123, what happens now.Do I have to restart the session, or can I do it without closing the session ? That becomes very important. Use a shell-based tool (like most CAD tools). I prefer Python, TCL is good as well.

Here's three cheers to brain-gain!!

[Python] Disassemble code, lambdas

2004-12-23T18:52:00.000+05:30

Here's two nice python features:

Lambda's are unnamed pseudo-functions, which are a functional programming tool

normal functions

def f (x): return x**2

Lambdas

f = lambda x: x**2

Usage(for both)

print g(8)

To disassemble python code and see their bytecode (python can also precompile your code ...somewhat like java)

>>import dis
>>dis.dis(f)
0 LOAD_CONST 1 (1)
3 LOAD_CONST 2 (2)
6 BINARY_ADD
7 RETURN_VALUE
>>dis.dis(g)
0 LOAD_CONST 1 (1)
3 LOAD_CONST 2 (2)
6 BINARY_ADD
7 RETURN_VALUE

[Business] ROI multiple for startups

2004-12-23T16:21:00.000+05:30

Ed Sim in his latest post about venture investing, writes about the average valuations of startups. The original source of his data - National Venture Capital Association has very interesting data about venture-backed mergers and acquisitions for Q3'04 . As compared to pre-bubble 1999, when average size of disclosed deal values was close to USD 231 million, the same parameter was USD 91 million for year ending Q3'04 .

In an earlier post, Ed Sim says that software companies need an average of USD 50 mn to create meaningful value and ripe for exit. At current exit valuations, that is an ROI multiple of 2.
I think that Ed needs to come to terms with the fact that ROI multiples of >4 will only be possibly in Fast Growing Developing Economies (FGDE). For example according to this presentation[ppt] of the Indian Venture Capital Association, average investment in an Indian startup amounted to approximately USD 18 mn. This year, IPO's of companies like TCS touched a high of USD 1.2 bn. The Indian IT market grew by 30 % in FY 2004 [source:Equitymaster], while China grew 45 %[source: China Daily].

If VC's expect a ROI anywhere close to the pre-bubble figures, then they should look up the next flights to Beijing or New Delhi.

[Business] ESOP expensing for corporates

2004-12-22T22:00:00.000+05:30

On Dec. 16, the US Govt. Financial Accounting Standards Board passed into law, a resolution that requires all corporate entity's to start expensing stock options, beginning mid-2005. This resolution applies to all corporate entities, with startups and private companies being given an extended deadline.
My favorite software Guru - Joel Spolsky, explains why expensing stock options is a good thing. In short, it boils down to this:

A company may choose to pay compensation in form of cash. Suppose this cuts liquid cash reserves by half. This means, the company's stock value is cut down by half. Now, instead of doing this, imagine the company doubling its stock (to distributing it among employees), this would halve the stock by half anyway.

However, now the company has cash on hand- cash which it can show in its annual report and claim as profit. Joel makes the case that companies should expense one million dollars. But I do not think it is accurate.
My argument consists of three points:

Suppose my current share is worth Rs. 1 this year and I issue one option at Rs.1 .What is my expense: 0

Now, next year if my share price rises to Rs.2, my expense is Rs.1. If however, my share price falls to 0.75, then my expense is still 0. So expensing stock options at issue, does not seem to be very effective.

Cash in hand is always better.

There must be some discount, on the basis of existing Reserve Bank interest rates, on the value of cash in hand. This is the least that you can do, since cash can always be re-invested, fuel growth and insulate against downturns. Your stock is worth zilch, if a startup had no cash to pay its ISP's after the first year.

The expense estimate calculation is not accurate

I'm sure this is an oft repeated concern. The Black-Scholes method for expensing options, is only directly applicable to negotiable instruments, with short vesting periods. ESOP's are not negotiable and have a gestation period of several years. The method has to have an added discount related to the strike-price of the stock and the vesting period.

Last mile broadband through Stratellites

2004-12-18T22:10:00.000+05:30

The latest update on the Globetel saga of providing last mile broadband access shows wireframe models of their stratellite. This being my favorite for the best bet for providing last mile access in untapped markets. This has the advantage of not having a capital heavy and more importantly, time consuming investment, in laying out underground cables.
The problem with conventional satellites is the impossibility of maintaining two-way communications through home antenna units. A stratellite, which orbits at around 10 miles, would be optimal for this purpose.
If we think about this concept in an emerging market like India, there are multiple problems that could surface:

In a place like Delhi, which are quite close to mountains, how do you stabilise the airship.
Would it be tethered or free-flying? I prefer the free-flying model, because it has all the advantages of the tethered variety with none of drawbacks (hold the line!!). I suppose it would have to carry onboard spare tanks of helium to account for the leakage.
How do I ever, in the name of all that is good, ever get on board to refuel/maintenance? I suppose I would have to use another blimp. But that does'nt give me reponse time of ~1 hour, which is what I would have to have to hook the big fish - corporate customers.

However, its all about execution. Execute , then improve.

iPhone??

2004-12-18T17:12:00.000+05:30

Russel Beattie hypothesises about the next Jobs' job- the iPhone . Personally, I feel that this may be a huge opportunity for small companies like Pixo( who made the OS for the iPod). Paul Mercer, the man behind Pixo has another startup called IVentor (not sure about the CamelCase!) . Their pitch is to make Java software for ... mobile phones.
I wont be surprised if he is already working for an OS for the iPhone.

Friction-less business

2004-12-17T22:17:00.000+05:30

Steve Brotman's post on removing friction as a viable business model was an eye-opener. But is removing human beings from the workplace, the general idea behind removing friction. I believe not.
See where Google is concerned and to a large extent Ebay (but not Amazon), is that there is almost no customer service. The biggest problem with customer service is ... service . To remove the human aspect from operations like this, lead to increased friction - in terms of goodwill.

Remove friction, insofar as customer experience is concerned - not at the cost of it.

Power of pricing

2004-12-17T22:00:00.000+05:30

My favorite software guru - Joel Spolsky dedicates his latest post on educating us mere mortals about how to..or rather how not to set a price to your software.

Maybe I should go through the terribly complex, but absolutely brilliant pricing-bible by Nagle.

NVidia's world domination

2004-12-17T20:46:00.000+05:30

One of the most intriguing, and exciting posts I read was a 2002 Wired magazine story about nVidia CEO -Jen Hsun Huang- planning to make nVidia , a contender for Intel's pie.
This year, NVidia has been on a roll, since its 6800 core GPU's outperformed ATI's x800 cores in Doom3 benchmarks (ATI does perform better on Half-Life, but that is due to ATI's 24 bit shaders which are used heavily by HL2 compared to nVidia's 8,16,32 bit shaders).

Today's story in EETimes adds a small footnote to those plans. However, with news of nVidia's cross-licencing pact with Intel, I wonder if it is hedging its bets.