In a recent report by Forrester Research, analyst Frank Gillett makes the case that HPC and grid computing are not generating broad interest and adoption in the enterprise. He comes to the conclusion that vendors should emphasize customer business solutions rather than technology themes. GRIDtoday editor Derrick Harris has already weighed in on the grid computing side of the story (see Why ‘Grid’ Doesn’t Sell). I offer my perspective here for the HPC case.
Forrester initiated global surveys of enterprise IT decision makers in 2007 to gauge their interest in using HPC and grid computing. From the results, Gillett concluded that “HPC garners more adoption and interest than grid, but not in a majority of firms.” This in itself is no big surprise; the broader enterprise market still dwarfs the HPC subset. And the fact that HPC does better than grid computing is a function of the former being better defined and more broadly established than the latter.
What was a bit surprising to me was how big that “minority” of interested HPC users actually was. Although it varied by region (43 percent in North America, 39 percent in Europe, and 50 percent in Asia/Pacific), overall nearly half of the people surveyed had some level of interest in the technology. Either they were currently using HPC; were planning to use it in the next twelve months; or were interested, but not currently planning to use it.
Of all those interested though, only about half were actually using or planning to use the technology. Gillett seems to be looking at this from the glass half-empty perspective, but if you’re a marketing weenie, you’re probably more interested in learning the whereabouts of those “interested” non-users. Maybe this just reflects the personality difference between analysts and marketers.
The more sobering aspect of Forrester’s results showed that, based on data collected in 2005 and 2006, adoption and interest is growing slowly, while awareness is growing rapidly. That also reflects the reality of the commercial HPC market. While IDC has seen steady, or even robust, growth for the last few years, I suspect a lot of that is from repeat customers who are scaling up their computing resources. Since HPC went mainstream, the real logjam in the customer base has been what the Council on Competitiveness has dubbed the “missing middle.” These are the potential HPC customers between entry-level and high-end users, who would benefit from modest-sized high performance systems if they knew how to apply the technology to their business.
Gillett’s prescription is to stop selling technology and start selling business solutions. HPC has a lot of street cred these days, but ultimately customers are interested in solving their problems, not learning your technology. The more mature HPC system vendors have figured this out and have built up in-house expertise in various industry segments (e.g., financial services, oil and gas, manufacturing, etc.) so they can speak the same language as their customers.
This alludes to the final problems of selling HPC: the meaning of the term itself. Is it high performance computing, high productivity computing, high-end computing, supercomputing, scientific/technical computing, data-intensive computing, ultrascale computing etc? No one knows. And customers really don’t care about computing taxonomies anyway. The more nebulous the term is, the worse it is for the seller. This is why terms like “virtualization” and “SOA” are headed for the marketing graveyard.
There is one very large HPC customer segment where Gillett’s conclusions might not apply: government and academia. It’s not clear if Forrester only surveyed commercial IT enterprise people or included non-commercial ones as well. Government and academic IT decision makers are culturally different than their commercial counterparts. They tend to be more interested in the HPC technology for its own sake and have a more general appreciation of how to apply it.
For example, purchasing a computer for a government lab or university that ends up on the TOP500 list can raise the prospects for contracts, attract more top tier people, and generally elevate the standing of the organization. And since a lot of decision makers in this segment are very savvy about computing, they would tend to be more receptive to a technology story that they know how to connect to their application. Of course, for organizations like government labs and supercomputing centers, HPC use is already at 100 percent.
Despite the lack of broad interest in high performance computing that Gillett describes, the market does appear to be growing rapidly. In 2007, the HPC server market (as defined by IDC) continued to outpace the rest of the industry. In February, IDC estimated that HPC server sales grew 15.5 percent in last year, reaching a record $11.6 billion — this despite a slowdown in the US economy. Meanwhile server sales growth in the entire industry was a meager 3.4 percent. IDC estimates that since 2002 the HPC market has more than doubled.
Combining the results from Forrester and IDC, a more two-dimensional view of the market emerges. While HPC use might only be broadening slowly, it appears to be deepening more rapidly. It’s difficult to say this definitively, since no one is tracking HPC expenditures at the granularity of the organization. But it does make a certain amount of sense. Unlike a lot of enterprise applications, which can run on virtualized platforms, HPC applications tend to be insatiable when it comes to hardware resources. Investing in things like more compute nodes, more storage, or faster interconnects often yields better results for the bottom line. Anyway you look at it, that’s a pretty vendor-friendly market.
As always, comments about HPCwire are welcomed and encouraged. Write to me, Michael Feldman, at firstname.lastname@example.org.