Waiting for the bat signal: the economy and the future of HPC

Print Friendly, PDF & Email

Doug Eadline posted the results of his Linux Magazine poll, “How has the recent economic events effected your HPC budget plans for 2009?” on the 16th, but somehow I didn’t see it until now. Here is what he found (out of 42 votes so, like my surveys, not exactly a large sample)

  • No effect (40%)
  • Pushing back major HPC acquisitions (19%)
  • Cutting back on people (14%)
  • Cutting back on maintenance hardware/software (12%)
  • Medium across the board cut back (10%)
  • Large across the board cut back (2%)
  • Small across the board cut back (2%)

Interesting, and to me the timbre of these results is that our sector is essentially ok right now. They seem to say that the actions that are being taken are more to prevent a major future crisis than to react to a current crisis. I don’t know Doug’s readers, but I suspect given that overall tone that they are from provider organizations — the labs and departments that buy supercomputers and employ the expertise that supports users doing business and science using HPC.

We know that the vendor portion of the HPC ecosystem is in distress. Advertising revenues from HPC sector companies are falling, and vendors are reducing workforce with thousands of jobs already cut, and more rumored in the works.  The HPC service provider organizations, at least those attached to federal dollars, seem to be ok right now, and Doug’s results sound to me like responses from those organizations. But I see a lot of potential for the distress to spread into the provider community, and I’m not confident that our current relative safety is an indicator that we are insulated from the problem.

In the industrial sector infrastructure investments like HPC are driven by bottom line analyses, and I expect that they aren’t thinking of themselves as “safe” at all. They know they are part of a larger business climate, and have to fit into the company’s overall strategy. In universities and the government, though, I hear quite different attitudes along the lines of “we are all insulated from this” and “that’s not our problem.”

In the university sector there is a longer lead time on the establishment of programs and expenditure of funds than the length of the current economic crisis, and so I wouldn’t expect to see a downturn in education that would impact staffing and purchases until the end of the Spring semester. As charitable giving and university endowments drop, so too will internally-funded research and infrastructure efforts of the type that often fund centralized computing efforts. And because of the lead time to build a pipeline on the other end, I expect this will continue well past the time that the broader economy starts to pick up.

In the government (and in programs funded by the government) the arguments are different, and require looking at individual departments. In Defense, for example, if Obama really does begin the process of disengagement in our overseas conflicts that many expect, then I would expect that DoD funding will begin to move to other departments which have been starved, relatively anyway, for funds over the past 7 years as we diverted funds to DoD. The military will likely try to protect what it sees as its core assets first — anything they can deploy in a conflict — which could mean that HPC budgets take another reduction as they did a few years ago (though that time it was to pay bills to support the war). On the other hand, Obama has pledged to increase science funding, and in general the mood these days seems to be swinging in that direction. So we may well see matching increases in computation funded through DARPA, NSF, NASA, and the civilian agencies.

Will they balance out? I’m not sure they will. Tax receipts will fall as more citizens are dislocated to lower paying jobs or are ejected from the workforce altogether. And the president-elect is talking about major expenditures in health care and our national infrastructure which will further divert funds. On the other hand, some of these efforts will undoubtedly create new opportunities for supercomputing (big new bridges will need to be simulated before construction, and comprehensive assessments of the state of our infrastructure may well invite large scale simulation as well).

HPC vendors are experiencing economic disruptions because razor thin margins and little access to working capital mean they are hypersensitive to even small changes in the market. But they could stabilize where they are now and slowly return back to the “critical, but stable” state they’ve declined into over the past decade. However, if the disruption pushes into the HPC provider community — the labs and departments that provide HPC cycles and expertise to users — and we see a large scale reduction in employment and acquisition, then I think we’ll be in for a wholesale restructuring of the  HPC market.

What could this look like? Think one or two companies for compute that are part of larger businesses with a very long term interest in remaining part of the market and willing (and able) to operate the HPC segment at a loss for 5-7 years. Maybe Intel, maybe IBM. I would also expect devastation in the software community, killing a lot of the companies that are doing some of the innovative things in programming models, workload management, and application debugging.

When you think about it, we really aren’t very far from this right now. Leadership can prevent this, leadership that will motivate DoD to maintain its HPC investments and argue successfully for pieces of the new streams of money that are likely to be routed into the new administration’s initiatives. Outside of one or two notable exceptions, this is leadership of the type we haven’t seen very much of in HPC for a long time. If you’re still out there, waiting for the bat signal, get rested over the holidays. And keep your eye on the sky: you’re about to get a call.


  1. […] West at InsideHPC.com covers a very important issue … that being the meltdown of the economy, and its impact upon HPC, specifically HPC […]