Windows Compute Cluster Server 2003 Released

grammar fascist writes "According to an Information Week article, on Friday Microsoft released Windows Compute Cluster Server 2003." From the article: "The software is Microsoft's first to run parallel HPC applications aimed at users working on complex computations... 'High-performance computing technology holds great potential for expanding opportunities... but until now it has been too expensive and too difficult for many people to use effectively,' said Bob Muglia, senior vice president of [Microsoft's] Server and Tools Business unit, in a statement."

Comments

Ok, I'll be the first

[Weaselmancer]

Imagine a Beowulf cluster of those.

Re:Ok, I'll be the first

[Ethan Allison]

But will it run Linux? (tm)

Too expensive my arse

[Umbral Blot]

Hasn't running "parallel HPC applications aimed at users working on complex computations" traditionally been done under Unix, and Linux as well. Seeing how Linux is free it's hard to see how "it has been too expensive", or "too difficult" (since unlike your home user the people running these systems are rocket scientists, I am sure a little command line use doesn't stump them).

Re:Too expensive my arse

[supun]

Your response is the same as mine.

When I worked a Motorola, and was part of their LUG, one of the members was talking about a Beowulf cluster they made. Like bad management, they ordered a bunch of desktop PC they couldn't use, and no one authorized their return. So they sat around in unopened boxes until his team decided to make a Beowulf cluster so they could model the electron flow around traces in an 8 layer circuit board before they had them actually pressed.

Each prototype board cost around $10,000 to create. And after that you have to test to make sure the electron field, around a trace, does not affect another trace. Manually it took a long time and is prone to errors. So if there is a problem, it's another $10,000, and another, until you get it correct.

With this Beowulf cluster they could model the electron flow around a trace and then only make one prototype, saving a ton of money and time. And this was all done with an ISO off the net and a bunch of forgotten computers.

Re:Too expensive my arse

[geoff lane]

You can build an HPC from random PCs but it will be crap because the PC to PC interconnects will be too slow. Real HPC needs highspeed, low latency internal interconnects and these are expensive. But I fail to see how paying a "Windows" tax will make matters cheaper, or easier.

Re:Too expensive my arse

[0racle]

Not to many are using Fedora or Slackware on some white box with parts from Best Buy to do HPC. They have been altered to specifically run on hardware that was made specifically for this, and even then management of it is not exactly simple. Not that I believe that 2003 Server will suddenly change that but just using Linux somewhere does not automatically make it the cheapest way.

And I believe the correct answer to your question is Traditionally it has been done by tuned versions of commercial Unices which added to the base cost of the OS over and above the very expensive custom built hardware. Recently Linux has become able to do many of these tasks by similarly being modified at a significant cost running on the same expensive custom hardware. The recent HPC installation using mostly off the shelf parts (they didn't use Ethernet) was the one at Virginia Tech and that ran OS X, not Linux.

Re:Too expensive my arse

[kylegordon]

You can still have real HPC with slow interconnects. It all depends on the application for the HPC. If your data has a high scatter rate that requires large amounts of data transfer all the time, then you need fast interconnects. On the other hand, if your data can be sent off to a node to be crunched on for 2 hours, then a bog standard gigabit ethernet interconnect will do you just fine.

Re:Too expensive my arse

[zcat_NZ]

When I first saw this on google news the headline was something about Windows for Supercomputers..

My first thought was "Oh, they've finally announced the real hardware requirements to run Vista"

Re:Too expensive my arse

[multimediavt]

Ummm...I know of several clusters on our campus (VT) that are made of white boxes running Fedora, Gentoo, or Suse. One is a 200 node (400 CPU) Opteron cluster with a Myrinet interconnect named Anantham, and built by Dr. Varadarajan's graduate students. There are other smaller clusters ( 16 - 32 nodes ) of various design that are running on GigE. All of them built of white boxes and other off-the-shelf components ordered from mail-order companies. In the case of Anantham, all parts were ordered separately, i.e., RAM, motherboards, processors, cases, etc., and the system integration was done onsite. So, when you say, "Not to [sic] many are using Fedora or Slackware on some white box with parts from Best Buy to do HPC," I'm guessing you are referring to those in the TOP500 List? If so, yes, there aren't many that submit to the TOP500 List (from large sites) that are using a non-commercial version of Linux, i.e., RHEL. Many of the larger sites are going through first tier vendors (Dell, HP, IBM) for a turnkey cluster solution, but they are paying a premium for those systems for the sake of time to production. They could just as well buy white boxes, but they would be spending a great deal of their own time weeding out problem nodes and components that could be better spent on doing science and supporting users. Academia can afford to take the time to do this, DOE labs cannot, although, that paradigm is quickly shifting as academic budgets tighten and competition in the Computational Science and Engineering arenas heats up among research institutions.

Clusters (the topic of this original post) are not "traditionally [...] done by tuned versions of commercial Unices [sic]". Clusters are traditionally built with off-the-shelf components with Linux and specialized APIs and drivers for the interconnect being used. If you want to talk about HPC BEFORE 1998, then you are looking at large monolithic systems of a custom built nature.

System X does have a GigE network, but it is primarily used for management and job startup within the cluster. We have had a few users with specific MPICH2 needs that have used the GigE network for message passing, but the GigE network was not designed for that task. Our primary communication fabric is IB. We are currently running Mac OS X (10.3.9) on the system, but are evaluating alternatives.

Re:Too expensive my arse

[Antique Geekmeister]

There are exceptions: some folks do wind up digging up racks of old servers, at rock bottom prices or even for free, as their data centers or deployed installations decommission them. You can inherit quite a lot of slightly outdated hardware this way: if you can justify the electrical expense of running them, they're quite convenient for massive, lengthy computing jobs.

A lot of cluster managers also mistake "really expensive, physically robust servers for "will always be working". The complexities of such setups and the general frequency of failure of "high availability" software itself means that the much vaunted 99.99% uptime of such systems is usually based on serious cooking of the numbers, not any metric actually used in the field. After the crops of failures of things like the old IBM deskstar drives, the run of bad tantalum capacitors in Dell motherboards, and other failures that strike entire classes of brand new hardware, it's often better to use older, cheaper, burned in hardware that's had the BIOS updates and the kinks worked out, and save the extra money for the next round of upgrades in six months or a year.

And for us mere mortals...

[Bananatree3]

"but until now it has been too expensive and too difficult for many people to use effectively"

and what about the site licence needed for this baby, huh? For us mere basement-cluster builders, there is a cheaper, open source alternative: The OSCAR Project ( Open Source Cluster Application Resources). Yes, it runs on Linux, but it is a nearly step-by-step system of setting up HPC-level clusters. It is being used on many 100+ CPU High Performance Clusters around the world, and it is free without those pesky site licences.

And in other news...

[nurhussein]

...the HPC community of scientists and engineers continue to not care.

The same folks who operate nuclear accelerators probably don't have that much of a problem operating computers that they need Clippy and pretty colours to help them out "in case they get confused".

Too expensive?

[onlysolution]

"...but until now it has been too expensive and too difficult for many people to use effectively..." According to their licensing model EVERY machine costs 469 dollars... Meaning a 20 machine cluster would have a 10,000 dollar overhead just on the OS alone. Not to mention the fact that you'd be compelled to buy it again as Longhorn Cluster Ed. in just a couple of years... It seems like a little work setting up a free OS cluster would be a vastly preferable option, is there really any need or reason for this (at this cost anyway)?

2003=2006?

[John_Booty]

It takes some serious marketing balls (and/or or a lack of marketing brains) to release a product branded "2003" when we're already halfway through 2006.

I actually have to applaud the naming move; it accurately lets everybody know that this product is based on Windows Server 2003. It would have been quite misleading if they'd passed it off as " Windows Compute Cluster Server 2006".

Wonder what the meetings between the marketing team and the engineering team were like for this one. :)

Absurdity can be profitable"

[asky]

"...but until now it has been too expensive and too difficult for many people to use effectively."

I know many people take exception to that remark. But not everyone knows how to build Beowulf clusters.Some of us thought it was insane when, in the 90s, Microsoft said they were going to enter the server market. Yet here they are. And who in their right minds would run their web services out of IIS? (Then again, Apache now runs on Windows.)

The point is, just because the idea is absurd doesn't mean it won't happen. If corporate consolidations put support for technical computing under the IT department, and support for Linux is considered toodifficult for the IT folks, it's only a matter of time before the decree to port technical computing applications to Windows.

The fact is, M$ has access to software vendors, hardware vendors, and large customers in ways that Linux companies do not. They can create markets where they shouldn't be justified (unless you think all operating systems really require anti-virus software).

I'd love to be wrong about this. But I've finally come to the conclusion that sound technical judgement does not stop absurdity from happening.

Another choice: Rocks Clusters

[joib]

Another, perhaps even more popular Linux cluster distro is Rocks Clusters.

While I don't have personal experience with OSCAR, Rocks is really good. These days, doing a cluster with a "normal" distro is insane. I think MS will have to think long and hard before they come up with something equally easy to install and manage as Rocks.

That being said, I think MS is not targeting Win CCS at academic supercomputing, which has a long history of using Unix/Linux, but rather they want to expand HPC to business customers who otherwise have a 100 % MS environment.

Missing the point a bit?

[nixascyborg]

The idea behind Windows clusters being 'cheaper' has nothing to do with the individual price of the OS (versus, for instance, free Linux); the named price is low, not free, but that is not the point of your savings with a Windows HPC cluster. The point is that most programmers work on a Windows platform and have experience with it. And if you program with/for Windows and, for instance VS 2005, MS counts on the effort of building programs that run on HPC to be considerably less effort than it is on a Linux (or Xgrid) cluster. Making existing Windows 'hits' clusterable (i heard mention somewhere of image, movie and 3d processing software) is easier because of this too; making it work on other clusters is a pain because there you would have to work in an environment you are not used too. Like all things with MS; they count on the familiarity and ease of use to make this all run. That is what makes it cheaper; you cannot get a Linux HPC programmer and if you find him/her he will be godawful expensive; for WinHPC it will just be 'another VS programmer' of which there are a lot. Look for MS to add testing, debugging and development aids for HPC in the upcoming versions of VS.

Re:The First???

[Anonymous Coward]

I did manage to flummox the professor with the following question (I'm being serious here!):

Why does a 256 or 512-node computational cluster need a copy of Windows Media Player on each machine?

It was then that I realized that MS's approach to operating systems targeted to different applications is not to strip anything out, but to add layer upon layer of extra functionality to their basic "home computing" OS.

Yes, friends, their Windows 2000 computational cluster software does indeed ship with Media Player.

I can't speak of SQL Server or other variants: does anyone have a clue?

Re:Absurdity can be profitable

[mikeb]

"But I've finally come to the conclusion that sound technical judgement does not stop absurdity from happening"

Something that the majority of Slashdot readers seem not to understand (and with justification) is that purchasing decisions are not rational.

A basic training course on sales techniques will, unless it's totally bogus, emphasise the fact that purchasing is based on emotion, not rationality. Some 80-90% of all sales are emotion-driven and then sometimes post-facto justified by selectively picking facts.

As the world becomes a more complex place and huge amounts of information become available to prospective purchasers there's a kind of paradox emerging that will horrify economists who cling to the theory that perfect markets are based on rational purchasers with perfect information, because the reverse is happening.

Most purchasers are not analytic personalities. People who hang around Slashdot underestimate how much they have (in general) honed their own analytic skills with years of practice while most middle-tier managers in corporates never did. For those non-analytic people, being asked to rationally evaluate a mass of facts and statistics is a SCARY proposition. That's not how they got their job, they did that by looking good in a suit and licking backsides more or less assiduously whilst being ok at judging how the politics are shaping up. Their skillset is way different from yours and they react differently.

The more information you make available to those people, the less they are likely to use and the more they will look around for 'safe' decisions. This will be especially true if their promotion prospects may depend on the outcome. THEY ARE NOT SPENDING THEIR OWN MONEY, it's the company's. Their decision will be based on the likelihood of retaining their job or getting promoted before their mistakes are discovered.

So, figure for yourself. On the one hand some technical guy they distrust because he's smart can 'download an ISO from the interweb and build a cluster myself' or 'buy from Microsoft'.

The first bit of irrational figuring will be 'the Microsoft stuff costs tens of thousands but the geek says it's free - that does not compute, he must be wrong'. The second will be 'if it goes wrong who will get the blame'. Guess the outcome of that one for yourself.

The result is fairly predictable IF you understand the parameters. Microsoft's marketing does understand where it's operating and will be well aware that its customer base is heavily loaded with irrational people. Most likely they are hearing squeals from that customer base asking where Microsoft's compute cluster solution is because 'we want to buy one'. It would be foolish not to give them one surely?

compliance to standards?

[WinEveryGame]

Per Microsoft document:

"Microsoft Message Passing Interface (MS-MPI) implementation is fully compatible with the reference MPICH2"

I guess given the fact that Microsoft is pathetically behind Linux when it comes to high performance computing, they may actually play by the rules here.

Anyone has an insight on this one? Do they have a API lock-in strategy here as well?

Heh!

[jd]

Microsoft delivered the keynote speech for the Supercomputer conference SC|05 in Seattle last year and had a huge stand in the main hallway. Most people were walking straight past it and were gathering round the stands offering free Linux CDs. (I would really like to see at least one of the BSDs to add something like bproc - Linux has some amazing capabilities and those can and are used in many applications, but NetBSD has better I/O throughput and there are many cases where cluster applications are I/O-bound rather than CPU-bound or feature-bound.)

Microsoft's MPI implementation is, if I understand their materials correctly, based on MPICH (a BSD-licensed Open Source product) with some in-house fine tuning. MPICH is a good reference implementation but is not terribly fast and is getting to be long in the tooth. Far as I know, it doesn't have much in the way of fault tolerance in it, either. LAMPI and OpenMPI are built for speed (although I've found OpenMPI has room for substantial improvement) and have some fault tolerance support. So, they don't seem to be using an amazing architecture.

Last, but by no means least, Microsoft's freebies were limited to an Opteron-specific Windows 2003 Cluster Edition beta and a cookie. By comparison, many others had booklets on what their products did, papers on the theoretical work being done, working demos (the molecular modeler with forced feedback was amazing) and some highly knowledgeable geeks to answer detailed technical questions.

Microsoft may - someday - be an interesting player in the cluster market. Right now, though, they really don't seem to get what it is all about. I'm not trying to bash Microsoft here, they really don't have a product that is useful for the high-performance market, and seem to have the wrong libraries and interfaces for using the servers in a load-balancing, fail-over or distributed storage environment. This isn't to say the other vendors were perfect - I saw many areas that were horribly inefficient and poorly implemented - but rather that Microsoft would have done better to have come back from the show and re-thought what it was that they wanted the Cluster Edition to do.

High-Performance Linux

[jd]

If OpenMOSIX is compiled into the kernel, the total effort required to set up a Linux cluster is virtually nil - you need to tell it what nodes there are in the cluster and it transparently takes care of the rest. A home user with ZERO programming experience but has two or more computers, a hub and a working knowledge of what an IP address is can configure a rudimentary cluster in under five minutes. It may not be optimal, it'll rely totally on OpenMOSIX to do the process migration, and without any apps that can take advantage of it, it would be a little pointless, but it could be done. It requires no expert knowledge or significant intelligence. If you can operate VI, you can operate a cluster at that level.

Difficulty, therefore, is NOT a significant factor in all of this. Ok, what about expense? Well, you're right that Linux is free. So is OpenMOSIX, OpenMPI (and many other MPI implementations), PVM (another messaging library), Lustre (a very high-performance network file system), many scientific and mathematical applications for clusters, etc. There are clustering patches for PoVRay, and it's always possible to write a script to have multiple machines render parts of images anyway. I'm sure there are other applications out there that I'm not thinking of right now, and it's only a matter of time before more "mundane" applications can take advantage of clustered environments. They already do, on Plan 9, to some degree. Oh, Plan 9 is also free.

Cost would appear not to be a major problem either, then. Optimizing is the only thing that is in any way difficult, and a GUI system that doesn't let you get to the really fine detail won't help there. More time, effort and money is spent on optimizing than on anything else, and I simply can't see any possible way that an OS that is designed for ease-of-use by hiding the intricacies can in any way help in that.

Cautiously optimistic

[hpcanswers]

I think Microsoft's reason for pushing into HPC is to provide better software development tools for clusters. Can you imagine being able to program in VB.net instead of C99? After all, physicists are there to do science, not write code. Plus, MATLAB (Distributed Computing Toolbox) and Mathematica (gridMathematica) will both be available for Windows CCS, and I imagine Star-P may be out before too long. All in all, I'm cautiously optimistic about getting better development environments available for supercomputing. Of course there is still the concern about license costs and the resource-hogging GUI.

I blogged about these topics a while back, both MS in HPC and better programming tools for supercomputing:

http://hpcanswers.com/plog/index.php?op=ViewArticl e&articleId=27&blogId=1
http://hpcanswers.com/plog/index.php?op=ViewArticl e&articleId=25&blogId=1

Re:The First???

[Bert64]

Does it also need a hard disk in every machine? Imagine all the extra heat produced by 500 superfluous hard drives. All the clusters i've built have been diskless, and booted from the network (or from a floppy, i did produce such a ghetto-cluster from old hardware once as an example)

Re:Absurdity can be profitable

[hackstraw]

...emphasise the fact that purchasing is based on emotion, not rationality. Some 80-90% of all sales are emotion-driven and then sometimes post-facto justified by selectively picking facts.

Maybe with things like cars and clothes, but clusters are merely machines to crunch numbers. Kinda like a big calculator, and little emotion goes into designing and using them. Its bang/buck. Thats it.

Microsoft _may_ be able to sell this HPC edition to some PHB out of emotion who is completely clueless and has clueless admins as well, but an OS has little to do with an HPC system. In fact, the less of the OS the better. Most of the time, HPC apps are in user land. The OS does basic memory management and I/O, but that is it.

Most all clusters are Linux. Why? Its good and cheap. You don't need the scalability and robustness of say Solaris, because you (typically, almost 100% of the time) only have one thread per processor. Yes, I know with large SMP machines, the OS can and does matter, but those rarely have the bang/buck ratio of clusters. The two big guys that have done this over the years (large SMP/NUMA/NUMAcc, etc) are SGI and Cray. And both of those companies are hard for cash right now. IBM probably does not make money, or much money off of their large number crunching systems, but they are probably viewed as RND, not a "for profit" good or service (I could be entirely wrong here regarding IBM, but thats my hunch).

I don't know what Microsoft is doing with this product. Like someone else said, its probably just a "me too!" thing. In looking at their "details", they do not mention using desktop machines at night. The is a BIG miss by them, because that would be one of the only things that could even make this a marketable item for an already primarily MS outfit.

The more I think about this, the more silly this sounds. Yeah, I'm an anti-MS guy, but I try to give them the benefit of the doubt, but this product seems completely worthless. Actually, now that I learned that this is an only 64bit offering, I believe this is a way for MS to sell a product for beta/stress testing of their 64bit server offerings.

To close this post, from the FAQ:

Q. How does a Windows-based compute cluster compare with a cluster running UNIX or Linux?
A. There is little substantive difference, but UNIX-based solutions should be fully ported to Windows to realize the full benefits of the Windows operating system. There are several differences between UNIX-based operating systems and Windows. For example, I/O operations and threading are different on UNIX-based systems than they are on Windows. I/O intensive applications will benefit from using Windows native I/O APIs rather than UNIX style I/O APIs.

Emphasis mine. The second bolded part is important. That porting is expensive and time consuming. Especially when its common for codes to be 30+ years old and designed for UNIX systems. Sounds like vendor lockin to me. Wow, typical Microsoft.

DUPE!

[x2A]

MS actually /did/ release it 3 years ago, but you know how slashdot is for posting old/dupe stories ;-)

Actually they don't but

[crovira]

now that super computing has been turned into clustering and there are lots of people doing it (like it hit >$x billion,) it has apppeared on microsoft's radar.

Unfortunately for Microsoft, the terrain's already covered by Linux and those systems are a moving target with cost-benefit lines that Microsoft CAN'T possibly over take. (The software is $-free and open source and the users WANT collaboration.)

Its a technological death trap for Microsoft. (I can just hear the SNAP. :-)