HPC in general has demonstrated time and again that it provides value in up and down markets. The real value of being able to get (even approximate) answers to “what-if” questions has not been accurately measured or accounted for. Moreover, much engineering and R&D; work depends critically upon simulation.
I expect companies to be far more frugal with new acquisitions, and want to focus upon getting more value and work out of their existing systems. Moreover, I expect that technologies and processes that lower barriers to getting computing cycles inexpensively will be on the upswing.
That is the bottom line. How this is going to unfold will be interesting to watch.
The market is going to get more competitive. Which means, for some segments, the only thing that will matter will be price. For other segments I expect to see people demand more for less. Competitors able to deliver similar or better results (performance and price) will win business. Weaker larger competitors will have no choice but to either adapt to the situation, or abandon that section of the market. Retreat to the high end is a tried and true strategy for eventual failure, especially as the market moves down-stream, and HPC shows up on desktops.
Universities may not curtail their spending as much, and may in fact take advantage of the more competitive market to increase their capacities. This would be true in areas with more grant money than others.
Engineering companies are, to a large extent, hurting right now. Downturns tend to impact them early, as people cut spending on new products, and on R&D; in companies. So they need to extend the life of their products, and get better value from them.
Financial companies are in somewhat of a strange situation. The ones that are stable are doing ok. The ones that aren’t are in a tail-spin. Both groups seem (from our discussions with numerous) wedded to various brand names. They are happy … or … have been happy … to pay a premium for a function. Now, I would argue, not so much. They have to be far more frugal than they have been in the past, and get better value for their money.
Seismic processing firms are busy. Their needs (processing, data storage, …) are exploding. This is not to say they will buy huge Crays. They are already quite frugal. They want the maximum performance at a minimum price.
Medical data processing (image processing, storage, informatics, …) are going through similar booms, but due to modality improvements.
Basically in all cases we see, users need to curtail their budgets. But their needs are expanding.
There was a comment on various groups being “stingy” in HPC. Ok, some groups do buy inappropriate-to-mission gear. Yes, we have seen this, sadly, up close and personal. And then when they fail, they don’t grasp why. Worthy of paraphrasing Einstein, “make it as cheap as possible, but no cheaper”.
This is why I think mixtures of accelerators and desktop or personal supercomputing are going to dominate HPC soon. Also, this is why we believe that JackRabbit and our forthcoming ΔV are so compelling. JackRabbit solves the problem of the highest performance you can get at a good price (well under $1000/TB with GB/s level disk and network bandwidth), and ΔV solves the problem of minimizing the price you pay per usable TB (a dv4-36T with 36 TB raw in 4U, 31.5 TB usable in RAID6 starts well under $15k). In addition, our consulting and re-engineering business appears to be on the upswing, as well as our support offerings being in higher demand. Companies have to make do with less, and get more out of it.
Cloud and remote supercomputing vendors like Amazon, Tsunamic Technologies, and others will likely gain from this as well. Their core concepts are to lower the per usage barriers to HPC. These are good models. They will work out well.
Some other vendors are going to be badly stressed, as their core business and market assumptions are rendered asunder and incompatible with the environment and climate they find themselves in.
I would not be surprised to see more bankruptcies.
This said, one of the more ominous things, which will impact all HPC hardware folks across the board, is the freezeout of the credit markets. If an HPC provider cannot get the capital they need to buy equipment/parts to build their systems, it almost doesn’t matter that they have backlog. Banks will not loan a business money on existing PO’s. They will loan some money (at least in the past) upon invoice. Which means that unless capital is available, quite a few vendors would have to dip into operational capital and possibly do layoffs, in order to be able to afford to build for customers, if they didn’t have lines of credit they could tap. New lines are hard to get, and old lines that are not fully utilized may not be available in their full amounts any more.
Couple to this, something we have seen, and this is injurious to all vendors. Late payment. Pricing models make various assumptions about payment terms. Floating loans longer than those terms pulls capital out of companies, and prevents them from pursuing more opportunities. We have terms set up (with discounts) to encourage early payment. If we have to front a loan longer than 15 days, there is a cost of money issue to deal with, and with the credit markets in nearly full panic mode, the costs of even short term capital are substantial. Which could push Net-X (X > 1) terms higher in price. This is sad, but likely necessary. If getting capital for 60 days to cover the cost of gear will result in a 5% additional charge to the builder, then we have the problem that this cost is going to be factored back into the customers bill. They may not want it, but there is a cost to every decision, including a decision to finance payment for Net-X terms.
I do expect to see some culling and consolidation. Non-HPC vendors (vendors for which HPC is a mere foot-note, not a majority of their business) may curtail their efforts to focus upon their core. Smaller players offering value may be acquired (no one has spoken to about that seriously in at least 10 months so don’t worry, I am not hinting at anything), and I expect this to occur fairly rapidly. Rackable picked up slicehost recently. Gives them an edge in simplified “cloud” provisioning. Storage folks have been seeing quite a bit of acquisition recently.
Whatever happens, we do indeed live in interesting times.
Which reminds me, I need to finish writing something.