The economy and HPC: part 1

John West at covers a very important issue … that being the meltdown of the economy, and its impact upon HPC, specifically HPC expenditure. Being on the vendor side of this I can offer my observations and make some suggestions.

First, allow me to note something that might not be too obvious to those in academia or government labs.
There is no credit in the market. There is effectively zero ability to borrow new capital to fund parts purchases for HPC vendors outside of their existing credit facilities.
What this means is that nice things like Net-30 and longer payment terms are going to be a distant memory for quite a few vendors going forward. Unless you (as a small company) are willing to freeze operational capital … e.g. the stuff you use to pay salaries, lights, rent, shipping, etc. and tie it up into one particular customers purchases, there is a strong trend of increasing pressure to pay in advance. Small firms can’t extend credit to large university/government organizations, if they aren’t getting credit extended to them.
And let me underscore that this lack of credit is, without a doubt, the single most dangerous element of our economic predicament going forward.
To date, the US government has tried buying assets from banks, investing in banks, lowering interest rates, and now printing money in order to increase liquidity. The banks, upon getting this cash, immediately …. sat on it. They would not make new loans.
As small business people will tell you, the last place you go looking for capital, are banks.
Sounds strange, right?
Banks are afraid … no, strike that … terrified … of bad mortgages. They likely have quite a few on their books. Which means they have to write these down as losses, and cover their losses with hard cash assets. The more they lose, the more they have to cover. The more money they get now, is money they are using to hedge against current and future defaults. I don’t agree with what they are doing, but I understand it. They are hoarding cash, and will continue to hoard cash, until they feel safe enough in their own assets/liabilities to loan again.
Which is, well, a little problem.
It is a problem much larger than the automotive industries problems. Though the current auto bailout is a direct consequence of no bank wanting to bet on GM/Chrysler going forward. The risk was too high that they would default due to the current market climate, and the precipitous drop in demand for their products. This isn’t a reflection of the product offerings quality, it is just a simple decision the banks made about the likelihood of being paid back soon.
Which meant that the large companies suddenly had little to no room on their lines of credit. Which meant that all spending came purely out of operational capital.
See above for what that means … Fourth paragraph from top, about Net-30 terms.
It also meant that the the autos started a slash and burn cost cutting once they realized that they burn through huge amounts of cash per month. Slash and burn gets us to HPC, IT, people, ….
It is easy to forget, when you are insulated from it, that we operate in an economy. That in this economy, there are things that flow … capital, value, etc. There are assets, there are liabilities. Cost centers are liabilities. These are things you minimize. Profit centers are assets. These are things you strengthen. When your work is partially, mostly, or entirely disconnected from this often you don’t remember what it is like to live in this world after a relatively short period of time.
Allow me to give you some imagery.
Imagine, if you will, that you have to go out and get grants, which could take from 1 hour to 6 months or more to get, and that without these grants, you would not be able to pay for your teams salary, or machines. And the grants, well, they are one offs.
So you have many oars in the water at once, pursuing a huge number of opportunities. Yes, you watch inferior groups get large grants due to some lower cost-of-research capability that they have, or the ability to produce results by the megaton. You watch superior groups swing for the fences and strike out most of the time (baseball analogy: means write elaborate, complex, bold, daring grants), but occasionally knock one out of the park at an appreciable fraction of the speed of light (baseball analogy: means win a competitive bid with an amazing bit of research proposal and initial results). And that keeps their supers humming along. And their people employed. For the most part.
You watch new entrants come along with really neat ideas, and get a few grants. You see heavily hyped groups with not so good ideas do the ‘flash in the pan‘ thing.
And with the economic turmoil afoot, as John mentioned, there will be less grant money.
Some of these groups are not going to survive, their funding requirements assume high burn rates.
Back to companies as companies, the notion of high cost for (perceived by marketeers) high value doesn’t work in a recession.
HPC has value. Huge value. One can point to study after study from the Council on Competitiveness through IDC, Gartner, and many others. HPC reduces numbers of prototypes for manufacturers, reduces development time and increases throughput and quality via simulation for engineering firms, enables product development and deployment for data analysis intensive systems including medical instruments, safety systems, etc.
The value in HPC is in part, due to the opportunity cost savings of what you don’t have to do because of it, of the quality and productivity increase you get from using it, and in part from the innate value that it provides.
HPC has been going down market for a long time. This is not a bad thing, it should be embraced, as it drives HPC into a much larger market. Which further feeds on the need for higher performance. Desktop, and laptops today include GPU chips capable of calculations that the worlds largest machines would find very difficult a decade ago. This power is available to applications which can be re-architected for it.
What I am trying to say, is that despite accounting people’s desire to classify the hardware, software, people, and services of HPC as cost centers, they really are not. They are cost savers. Productivity increasers. Value adders.
Of course, it is hard to convince accounting of this.
So there will be pressure to reduce expenditures.
And this will impact the market.
Weaker providers will probably not survive. I have my own personal list of whom I think will be gone this time next year (chapter 11 or 7), and who will be in a world of hurt. I have an idea of whom will thrive and survive.
Business spending has largely ground to a halt. Even profitable businesses are betting that this downturn is going to be protracted and painful.
But the HPC users still need to do HPC.
Because that helps keep the costs down.
So they have to spend on something in order to keep these larger costs down.
But they have to keep costs down, so they have to spend as little as possible.
Somewhat of a catch-22.
John points out that at some point, grants are going to be impacted. Federal and state tax revenues are falling off sharply. Few states are in the black, most are in the red, some massively so. State funded projects are likely to evaporate.
Business spending has dried up due to lack of credit.
Business tax revenues are drying up due to lack of revenue.
Personal tax revenues are tanking due to the layoffs.
Second order effects, such as defaults on mortgages are on the upswing due to the layoffs, which further exacerbate the situation. This is unfortunately causing upticks in the rate of layoffs to conserve capital, and a further freezing of capital markets.
I know many good HPCers out there, looking for work right now, caught in one of the many RIFs (reduction in force, or redundancy for those speaking proper English).
Doom and gloom.
Is there a way out?
I think so, and it isn’t actually all that hard to do. Though it does require some … er … fortitude …
Curiously enough, I think HPC can also show the way. Doug Eadline’s survey makes me believe that there is.

Doug Eadline's micropoll on Linux Magazine's website
Doug Eadline's micropoll on Linux Magazine's website

Next post for this though … this is getting long.