There are some fabulous anti-OSS observations in this article - namely that if you know your shit you don't have to pay for an operating system with OSS.

Before I get into the HPC bit (and get flamed to death for what I said above), I'd like to point out that financial success in OSS is never going to come from big business. Unfortunately that crushes the dreams of lots of the Open Saucers but they've never had a good grasp on business. Big money it tech only comes from mass adoption and by and large OSS can't get it done.

Back to HPC. In my time at ORNL (Oak Ridge National Laboratory) the "Leadership Computing" people all bitched constantly that they had to hire some "super geek" (this coming from uber-geeks) to develop the programs and handle administration during their (rare) computing privileges. Most of the HPC people are advanced grad students who are working off of grants and every dime is important. Many of them feel that their limited funds are being wasted on finding and recruiting Linux people. The work they are doing might make them famous or rich, but the work has to be done first; they think that weirdo computer crap is harming their chances of fame/wealth.

If Windows can get into the HPC market with just one or two wins then they'll be made. Linux will quickly be replaced by Windows in the market and unbelievably be cheaper for the supercomputer owners. They'll only have to purchase the licenses once and since computing time is such a rare commodity it works out cheaper than hiring a battalion of OSS guys that never seem to agree on anything once they are grouped together.

Windows isn't getting into HPC on a lark. They've been getting requests at the highest levels of government and science for years you know.

(which seems to be the way things are going) isn't that going to pull the rug out from under the supercomputer market.

Or will spooks and nuke designers underpin it?

"The work they are doing might make them famous or rich, but the work has to be done first; they think that weirdo computer crap is harming their chances of fame/wealth."

These uber-geek grad students with small grants who don't want to know about Unix (whether that Unix is "Linux" is beside the point, it could be Unicos something even more mysterious) but who are uber-geeks who want to do HPC computing to be rich/famous are indeed an interesting breed, Realistically, they would be better served by a pool of in-house consultants who would help them get their application onto the big iron in the Cool Room than by shelling out for some pseudo-HPC cluster under their desk running Windows which they will be unable to manage or use effectively. The least of the things that will happen is that they will just forget to do the backup. I know the type.

I cannot see the adoption of Windows in any Academic HPC environment. Here (in our UK University site) we are lucky if the apps customers submit are written in fortran - many are much older than this. And what about those obscure libraries - FreeBUGs anybody?

In a nice clean, rich, oil exploration company maybe everybody has upgraded since 1989, but experience shows this isn't the case at the research end - and it is here that the next generation of users and supporters are made.

@Solomon Grundy

You said "Linux will quickly be replaced by Windows in the market and unbelievably be cheaper for the supercomputer owners. They'll only have to purchase the licenses once ..."

They'll purchase the licenses for HPC2008 and then a couple of years down the line, when HPC 2008 has been tweaked and released as HPC 20## they will have to repurchase the licenses again. Look at Microsofts history and you will see that certain bugs in their software are only fixed in the next version! It is one of the ways that they ensure they have a continual income. The HPC customers will need to upgrade or risk have incorrect calculations/

"Windows isn't getting into HPC on a lark. They've been getting requests at the highest levels of government and science for years you know."

Prove it, or you are just another Microsoft fanboy spouting bullshit.

At the end of the day someone is going to need to administer these clusters, whether they are running Unix, Linux or Windows. Are you trying to state that these people are going to risk all of their results by adminstering the systems themselves? I don't think so. They will end up hiring someone to do it. Academics are not extremely intelligent people who know everything. Most of them can barely use a computer!

"In many cases, these HPC sites know a lot more about Linux and tuning it for HPC than Red Hat, Novell, and other commercial Linux suppliers."

Given their history on the desktop and server, I really can't see this being any different for an HPC version of Windows - you're still going to need to hire someone to actually support it. And that person's still going to be an HPC orientated person (or not very good at their job).

Unless you're the kind of big shop that gets the proper support deals from MS. But, in that case, I'd have thought you'd be equally able to find and hire someone to run whatever other OS you might decide you want.

Now you can have malware from top to bottom.

"Windows isn't getting into HPC on a lark. They've been getting requests at the highest levels of government and science for years you know."

Yeah, ever since the beancounters took over running the asylum and most of those who actually understand how much processing power etc is wasted driving M$ bloat decided long-term employment and happy Pointy-Headed Bosses is better than sky-high blood pressure and constant headaches from hitting brick walls...

AC cos I'm one of the latter!

In my experince of the HPC crowd, most of the HPC people are people who have been grad students, and whille they were grad students they learned unix/linux and learned to love it.

The thing they actually bitch about isn't the need for super-geeks who do the optimisations and administration etc. it's how to get their programs to run faster, so work churns through the queues faster and they actually get to do their work (rather than sitting at their desks, reading the reg while stuff waits/compiles).

My guess is that the windows for HPC (like the Cray X1) will make a couple of big sales and headlines, then people will figure out that if they wipe the system and put linux on it then their queues run 10% (or knowing Redmond, 2x or 3x faster) and they'll switch in a flash.

Where the hell is this linux uber geek stuff coming from? And companies needing an in-house team with better HPC skills than redhat and novell? Utter bollox. Take a look through the debian, redhat etc. repositories, half the stuff in there is scientific. Why? Because the 'ubergeeks' doing this stuff know it makes sense to keep it in the open so they don't all have to re-invent the bloody wheel every time they need to get something done.

So maybe the beancounters get there way and put this overpriced bittybox system on $250000 worth of hardware, what then? A piece of software rarely contains everything it needs to run, there is usually a chain of drivers, libraries, external programs etc. For that kind of money your going to expect every one of these components to do their job in the minimum number of cycles, how are you going to do that with closed source software thats intended to operate on any kind of x86 hardware? It's not going to work. Do the sums, that $250000 system is going to be wasting at least 25% of it's cycles whereas optimised software can run with the same performance on a system costing less than half the price. Add to that the money saved by using a free OS (BTW the price mentioned for windows HPC is for a limited number of processors) and pretty soon you are looking at a situation where the cheaper system plus the developers to get the most out of it costs a fraction of the price of the heavy duty system, the inefficient propriotory software and a few clueless administrators giving the same performance (when it's running).

It's a shame MS's plan to dumb down computers and spread bloat is coming to the world of supercomputers.

I've had the pleasure of developing software in Windows environments, before working in HPC on UNIX systems as I do now -- and to me several things are clear.

1) @Solomon -- perhaps you were born yesterday, rather than Monday? A significant portion of HPC developers are UNIX/Linux geeks of a sort, and perfectly capable of administrative tasks. Large installations do have admin teams, as you would with any sizeable system..

2) Microsoft is not going to be able to touch the high-end stuff in the scientific HPC community with a barge pole. Quite simply, the OS is not geared for it. It's taken over a decade for Linux to become competitve on the desktop; I suspect for Windows it may take equally as long.

3) Engineers would go for small-end HPC platforms running Windows; many of those I know work solely on Windows platforms, running software such as SolidEdge and CFX; a 16-core Windows HPC server would be a gift to them.

It's not a completely daft idea; however, I hope Microsoft are sensible enough /not/ to market this as a solution for high-end applications.

A couple of points of information that could help inform the discussion:

1) Microsoft pricing for Academic customers is very reasonable and the per system licensing cost is a tiny fraction of the numbers mentioned in the original article. In the academic market, Windows is about equal in price to Linux.

2) Windows is performance competitive with Linux, in fact the highest efficiency top 500 system is currently at #40 coming in at 85% of peak and running Windows Server 2008.

3) HPC uses in the Financial services market go beyond derivative pricing, and the biggest consumers of CPU cycles are stochastic models that provide a probability estimate of future value. This occurs in Insurance, Banking and Capital Markets.

Full Disclosure, a significant portion of my income comes from Microsoft products.