Friday, May 29, 2009

Sane Advice and Good Lessons

I wanted to keep this for future reference (hence the content and not the link)


I’ve been an engineer for a while now, and have done a lot of things, but nothing else that I’ve done as an engineer has been as rewarding as building SiCortex with John Mucci and Jud Leonard. We started in a basement office and built the organization to 70+ employees and a growing number of installed systems.

SiCortex closed down engineering operations today, reducing staff to a support team that will service the installed systems.

While SiCortex’ demise will no doubt be attributed to everything from the alignment of the stars to the deplorable state of education in modern America, I’d rather write about the positive side. The SiCortex experiment demonstrated some things pretty clearly:

First, it is possible for a small company to compete in the computer systems business. There are some who will say that nobody can compete against “commodity manufacturers.” Ignore them: they are either shills for “commodity manufacturers” or they just don’t understand the landscape. There are only two true commodities in the computer business: DRAMs and wafer area. Everybody pretty much pays the same price for DRAMs. Wafer area is what you make of it. If you insist on building giant 100W chips, life will be tough. But if you use the silicon wafer area for something new, different, and efficient, a market will open up to you.

“Commodity” is a codeword for “x86” in much of the industry. In seven years of talking to end users and system purchasers, the non-x86 nature of the machine rarely presented much of an obstacle. In a large part of the high performance technical market the customers either own their own codes or build their software infrastructure from community owned sources. Yes there are a few major ISV codes, but -- surprisingly -- the available market outside of those areas is quite large (many billions of dollars). And if the architecture is attractive and there is sufficient market opportunity, ISVs will port codes (often for a fee.) SiCortex didn’t fail because of the x86 instruction set. There were a few prospects that shied away because of instruction set issues, but these were few and far between.

The SiCortex vision of making affordable high performance communications a primary focus of a dense cluster design created a new kind of cluster that couldn’t be created by the PC manufacturers. As a result, there are some places where SiCortex clusters are extremely well suited to the computing task. The SiCortex architecture was able to gain a foothold in power constrained applications (where either cooling capacity or electricity costs are primary concerns) and in applications whose run time is dominated by the cost of inter-node communications. The primary liability in the first generation product was its relatively weak memory bandwidth per processor. The engineering team was addressing that problem in the Gen II system that was under development until very recently.

Second, it is possible for a very small group of people to do amazing things and have fun while they at the same time. The first generation SiCortex systems were designed and assembled by a team of fewer than 40 engineers. The software team was tiny by system company standards. The chip team was fewer than 25 people. Chips done at the “commodity” manufacturers have more than 25 managers working on a project. During the first generation development at SiCortex there were no managers who didn’t have some other “non management” task. And it was fun. Even a bad day working in a team where everybody was dedicated to reaching the goal was more fun than working for BigCo.

Third, there is a need in the computer industry for more ventures like SiCortex. My greatest fear is that somebody else will go out and try to raise money from the Venture Capital market and be turned away with comments like “SiCortex tried to compete and they failed: so will you.” That is baloney. SiCortex failed for the same reasons lots of businesses fail: they ran out of money. The reasons had nothing to do with the product concept and everything to do with the execution and timing: May of 2009 is not a good time to be raising money: success takes a mix of luck, skill, and determination that just didn’t come together this time around.

The fact is, innovation carries with it a risk of failure. Large organizations may be incapable of creating an environment for genuine risk taking. The reward system makes chasing small market segments unappealing, and ambitious competent managers and leaders hear the message: aim for the big wins with giant market opportunities. Unfortunately this has led to an industry where uniformity is the dominant theme. But there are lots of computing tasks that don’t look anything like Excel or a video game. The industry needs daring experiments. They fill in the gaps left by the uniform copy-exact giants.

Companies like SiCortex pose no real threat to the BigCos of the world. But they can serve as an irritant: a reminder that the dominant model may not hold the ultimate answer. When SiCortex started, nobody was talking about power efficiency in high performance technical computing: the assumption was that BigCo was doing about as well as could be done. Now system vendors are engaged in a competition to see who can make the most outrageous claim to green-ness. (My favorite: turning down the fan speed so the chips run hotter but less power is spent on turning fans. That’s what passes for daring innovation at BigCo.)

I’d like to thank every one of my colleagues at SiCortex, and especially those who I recruited to the effort. Your faith in the vision, your good humor, your devotion to the cause, and your support through all the challenging times has been a gift to me. For those who are leaving SiCortex today, thank you. For those who are staying to support the ongoing efforts, thank you.

I am also grateful to all the SiCortex spouses, SOs, and families. The late hours, the screwed up vacation plans, the lost weekends, they were the price of admission. Thank you for all that you put up with.

And thank you, investors both big and small. Along with the major shareholders -- Flagship, Polaris, Prism, JK&B, and Chevron, there were angels and other stockholders who had faith in the vision and courage enough to put money behind it. The VC investors bucked a trend back in 2002 when most of the big money was saying that innovation in the computer industry was over. We showed them that it was not.

As an engineer, I’d spent most of my career before SiCortex far removed from the sales force. Working with the SiCortex sales team has been wonderful. For you technical geeks out there, if you get an opportunity to work with real sales people -- do it. Not only will you learn more about your own product, you’ll learn a lot about how people make the crucial buying decision. Sales folks are a special breed -- no, they are a whole collection of special breeds -- and the good ones work in a landscape that is as far removed from spreadsheets and mathematical models as ballroom dancing is from hardware design.

And for the customers, prospects, technical hosts, conference organizers, workshop coordinators, journalists, students, and educators who offered hospitality, advice, and encouragement, thank you. The last few years have been tremendous fun and often very enlightening. I owe a special debt of gratitude to the folks in the seismic exploration business who patiently explained the peculiar nature of their very interesting computational problems.

And of course, thank you to every manager, decision maker, geek, and visionary who made the decision to buy a SiCortex system. Thank you for your faith in the product and your willingness to join in our experiment.

SiCortex has been an interesting ride. There is still an opportunity for a buyer to come along, and I hope someone does. The current generation product still has a great deal of untapped promise -- only recently it has shown us some new tricks. The next generation builds on the concepts from the first. Whether SiCortex builds it, or some ambitious and determined new venture gets created somewhere, there is lots more to do in the computer industry beyond duct-taping desktop PCs together and turning off the lights to keep the bulbs from getting hot.

Primary takeaway: I’m glad I walked away from a perfectly good job with a very large company back in 2002 -- for all the bumps in the road, SiCortex was far more than an interesting ride, it was a thrill and a joy.

Dream a dream. Chase a vision. Build something. Repeat.

---Matt Reilly

1 comment:

Jason said...

Thank you for sharing your honest inputs :)
Lots of learning follows ...