|The Leading Source for Global News and Information Covering the Ecosystem of High Productivity Computing / April 14, 2006|
With their "Adaptive Supercomputing" announcement, Cray has taken a bold "catch us if you can" marketing position, and maybe they have redefined the game to some extent. After all, we all want a system that meets a wide variety of needs in a handy way. Cray's announcement got me thinking about the history of all-purpose tools. The kind of stuff Ron Popeil made so famous during the wee hours of pre-cable television. Baby boomers will recall staying up late to catch some campy old sci-fi movie, a Marx brothers classic or a really bad example of what passed for horror in the days of analog special effects. The "call right now" commercial promised fabulous results and undreamed of convenience, invariably yours for 19.99. And they made a fortune. For proof, go to any yard sale and count the 19.99 stuff on the tables.
But the idea of one-style-fits-all or "software first," platform design has a less glorious past. The promise of Universal UNIX quickly deteriorated into serious kernel modifications at every street address, and we don't hear anyone touting processors designed "from Linux up." I still recall reading Tracy Kidder's classic book "The Soul of a New Machine" about DG [Data General Corporation] and being stunned at the end when he casually said the completed new hardware was rolled into a lab so that an operating system could be written for it. It seemed backwards to me then, and it still does today.
Perhaps the notion of various processor technologies sharing the same operating system environment and living in the same cluster is good evidence of technical Darwinism. It might signal the death of standalone special purpose architectures and herald a new age of modularity for HPC. Or, maybe not. Is an all-Linux platform heterogeneous enough to make it happen? Selecting what blade gets the job based on the application does not seem like a major coup in and of itself. Making a truly integrated cluster on which several operating systems are dynamically invoked as needed to run various new or legacy codes would be a dandy piece of work. But that approaches science fiction, and the market does not seem to be crying out for it. Besides, producing tools to monitor and maintain the performance of such systems would be Nobel Prize material for sure.
The key for Cray, now that the claims are out of the bag, is time-to-market and proof of concept. If the adaptive architecture is robust and scalable, Cray has a strong story. But the window of opportunity for true leadership is very short these days. How long will each of the three phases of implementation actually take? That remains to be seen. However, by announcing their adaptive strategy now, they may have changed the nature of the game for some other vendors. And that's what leadership is all about. Now we'll find out if their marketing decision to position themselves once again in a leadership role is matched by their ability to outrun nimble competitors who buy into the adaptive concept. We'll quickly discover if the idea has enough gravity to motivate other vendors. If major customers delay decisions until they know more about Cray's new approach, we'll soon hear from a bevy of vendors, talking about their plans, claiming that their grandparents actually came form Adaptavia, that they spoke fluent adaptavian as children, and that they know all about it.
With the mind-blowing complexity of HPC applications and the ubiquitous use of multi-core platforms, it seems to me that operating systems, compilers and tools are perhaps as important as new architectures, adaptive or not. If adaptive systems make the best tools and applications available for all users all the time, Cray has grabbed the brass ring. Does Cray have, as their press release claims, a substantial lead in making adaptive clusters "because the company's depth of expertise and intellectual property encompasses the various processing types and their related compilers"? Maybe they do. Cray is a serious company, not so often given to over-the-top marketing hype. But, with all deference to the venerable engineering behind the Veg-O-Matic and the Pocket Fisherman, it was Ron Popeil's marketing savvy that made them icons of sales success. In any case, as a marketer, my hat's off to Cray for boldly establishing another leadership position, getting big buzz in the HPC community, and boldly crying "catch us if you can."
Bob Feldman has been helping advanced technology companies improve sales and marketing results for over 25 years. As President of HPC Marketing (www.hpcmarketing.com), Bob works closely with management, sales and marketing teams, providing objective insight gained from working with dozens of companies through good times and bad. His past clients such as HP, Intel, Etnus, National Semiconductor, Iona, Thinking Machines, Multiflow, Sequent/IBM, Ascend and Lucent represent a wide variety of technologies ranging from supercomputers to middleware, storage, data mining, simulation, software development tools, product testing and more. Bob has been honored with an Emmy Nomination for broadcast TV, four Telly Awards for excellence in industrial media, and a White House Assistantship for writing. He holds a BA in Communications and a Masters in Research Design. He can be contacted at email@example.com.