Windows Cluster Edition

eth8686 writes "Microsoft is aiming to have its first cluster version of Windows ready in time for a supercomputing conference this fall." From the article: "The next version of the Compute Cluster edition will extend to Microsoft's .Net programming infrastructure, letting developers write software using the C# programming language, he said."

Comments

What is the point?

[BWJones]

What is the fundamental difference with the "cluster version" of Windows? OS X clusters just fine and there are no "special editions" other than a few software additions that hardly count as a different OS. And Linux requires very little to get it in a cluster compute configuration.

However, Theimer said the cluster version will include some restrictions on how the version can be used to prevent companies from performing standard Web hosting or other functions.

Wow. When you compare this to the standard capability of OS X, it seems like a real rip off. You get reduced functionality. Why?

The first version will reproduce many basic features of Linux clusters, Theimer said.

Then why not use Linux?

The next version of the Compute Cluster edition will extend to Microsoft's .Net programming infrastructure, letting developers write software using the C# programming language, he said.

Ah, I see why now. But what impetus is there to use the first version if this is coming in the second version? Kinda like Windows 1.0 I guess.

Although such code runs more slowly than C programs running directly on Windows,

Aauuummm........

writing programs in C# that run atop .Net is easier and more secure.

Says who? It certainly is/will be easier but more secure is something that has yet to be proven. To date, the track record is not impressive.

Often, Theimer said, it's more important to have a program as soon as possible than to have it running at peak performance, he said.

Ah, the fast food approach to software design. Don't you know that stuff makes you code obese and causes an early demise necessitating frequent checkups?

A third version will include developer improvements to ease programming on clusters. It also will include high-level management tools and will help customers integrate their high-performance computing equipment with the rest of their infrastructure, he said.

This is going to be in the third version of the release? I guess they have been looking at Xgrid, Pooch and other software and it will take them two versions to integrate what others have already got.

Seriously, Microsoft. Please come up with some innovative features and give us something that no other vendor offers or in a package so slick that we cannot help ourselves, but to purchase the Microsoft solution. This is nothing that is not offered elsewhere in the market, but has the appearance of locking us further into a Microsoft paradigm.

You guys have the right idea in that cluster computing is going to be a bigger market than it currently is, but you have to be more hungry and learn again how to ship software that creates desire and meets your customers needs in a timely fashion.

Re:What is the point?

[Coryoth]

C has a really crappy track record of being secure actually. As does C++. Fundamentally, they are just fine. In practice, just about ever buffer overflow exploit around was enabled at least partly because the developers were sloppy and used unchecked buffers. This is not possible in C# or other .NET langauges.

There is a big point that is being missed here though. We're talking about cluster computing, presumably on a large scale. You don't bother with something like that unless performance is a top priority. Should security be a second priority? In terms of the code written to do computations on a large powerful supercomputer... hell yes! You see, if you have a huge expensive compute engine, you don't go randomly hooking it up to the internet, nor letting anyone who hasn't had careful screening get access to the damn thing. Security for a cluster needs to be strong - so strong that by the time you actually have any access to the cluster you can be assumed to be trusted and not worry about buffer overflows. Basically, if a cracker has account access to the machine to be able to use a local buffer overflow, your security has already failed big time, and him getting root is, by the point, the least of your worries.

Jedidiah.

Re:What is the point?

[hey!]

writing programs in C# that run atop .Net is easier and more secure.

Says who? It certainly is/will be easier but more secure is something that has yet to be proven. To date, the track record is not impressive.

Well, sounds like you're spreading FUD to me. Compared to Win32, CLR almost has to be an improvement.Consider that Win32 was designed pre-network, had networking added as an afterwards but before security was understood to be an issue, then has had security issues patched over the years as they arose. It's got to be a nightmare.

CLR has pretty much what you'd want to have designed in security-wise, designed in from the ground up. Stuff like having the JIT compiler examine intermediate code for type correctness; allowing code to require the principal to have certain authorizations to run; cryptographic authentication of code identity etc. Not to say there aren't going to be issues in the future, but almost certainly a system designed from the ground up with modern security ideas going to be better than the patched monstrosity that is Windows.

Often, Theimer said, it's more important to have a program as soon as possible than to have it running at peak performance, he said.

Ah, the fast food approach to software design. Don't you know that stuff makes you code obese and causes an early demise necessitating frequent checkups?

Well, you're just being deliberately obtuse here. What he's saying is that in a lot of cases where you need serious computing power, you're trying to come up with an answer. The right metric for this is time-to-answer. Time-to-run is obviously a big part of this for those kinds of problems, but if you can cut down on programming time by working with a better environment enough, a linear factor increase in runtime may be acceptable. It is actually more than acceptable if you can use the time you gain by working with a better programming environment to engineer a solution with a more efficient algorithm.

What is interesting is the implicit acknowledgement that Win32 is insecure and hard to program well for.

Then why not use Linux?

Well, yeah. It's going to be a tough sell right out of the gate, since people in this market presumably are more likely to know what they're doing. You can't fault 'em for trying.



Seriously, Microsoft. Please come up with some innovative features and give us something that no other vendor offers or in a package so slick that we cannot help ourselves, but to purchase the Microsoft solution. This is nothing that is not offered elsewhere in the market, but has the appearance of locking us further into a Microsoft paradigm.


Why? Why shouldn't somebody try to make a better product that embodies proven and well established ideas?

Is Linux technologically innovative? Well, no, not that much. It doesn't mean it isn't a good operating system, or innovative in other ways. But with a few exceptions just about all of the ideas and many of the technologies in Linux originate elsewhere. Nobody with any technological maturity considers it "cheating", it's really just good engineering to solve known problems with proven approaches.

Re:What is the point?

[nkh]

C has a really crappy track record of being secure actually.

Sorry for asking such a stupid question but do you really need security for a cluster? If it's supposed to be a big calculator used for scientific purpose, you don't connect it anywhere. Just write you code heavily optimized in C and assembly language, connect everything using sockets or MPI and run it. If you need a front-end like a web server, it's not meant to be fast and you can write this in a more secure but slower language. Where am I wrong?

cost

[TedCheshireAcad]

If you're spending $millions on a cluster, it's more useful to spend the money you'd spend licensing MS software on more computers for your cluster.

cost benefit analysis.

Re:Slogan

[A beautiful mind]

Sadly, this is the point.

Let me be the first one to say: Windows isn't

• Designed
• Ment
• Capable

for/of running on a Top500 server.

The most important part is the design on those systems. They need flexibility. Windows is anything but flexible. No wonder that the top500 is mostly made up from unix/linux systems.

They need customized things, not a toy. The people running those supercomputers want to customize things themselves. Windows is just not ment for anything else than desktops, thats the truth and i know i'll get flamed for it.

Not happening

[MetricT]

I work in the field (sysadmin for a 800 node cluster), and this is pretty laughable. Microsoft is desperate for the "street cred" of being able to handle high performance computing. Sun, IBM, Dell, HP, Apple all have it. Microsoft doesn't.

If they want so much as the proverbial foot in the door, they must 1) release all (as in *ALL*) of the source code under a GPL or BSD license, 2) make it available for free to all comers, 3) have user's 3rd-party apps (ISE-TCAD, CFDRC, etc) ported, and 4) provide a knowledge base equal to (All Linux + BSD hackers) * Google.

And that only gets their foot in the door.

Re:Slogan

[ackthpt]

The most important part is the design on those systems. They need flexibility. Windows is anything but flexible. No wonder that the top500 is mostly made up from unix/linux systems.

No argument there.

They need customized things, not a toy. The people running those supercomputers want to customize things themselves. Windows is just not ment for anything else than desktops, thats the truth and i know i'll get flamed for it.

Windows is the last operating system I'd associate with 'super computer', in any interpretation of the phrase. It's a good jack-of-all-trades platform, but I can't see running bloated code, particularly using the CLR. Maybe it's a completely different operation system than we see, as in 'only the kernel', without all the plug-and-play, DRM, and annoying as hell code which throws requestors up while your typing (to steal keystrokes and disappear to do The Bob knows what with your inadvertent instruction.)

Not true anymore...

[benhocking]

You see, if you have a huge expensive compute engine, you don't go randomly hooking it up to the internet, nor letting anyone who hasn't had careful screening get access to the damn thing.

Cluster != huge expensive computer engine. With software like clusterKNOPPIX (I was just playing with this today), it's really easy to take all of the computers in a research lab (that are already connected to the internet) and turn them into a load-sharing cluster. This is different from a super-computer, although you can presumably get some of the benefit if you are also running MPI (I haven't tried that locally yet). In case you don't understant the purpose of such a cluster it's so that when I want to launch 100 simulations (say to do a parameter sweep, which is embarrassingly parallel), I can launch them all from my local computer and openMosix will automagically distribute the workload across all computers in our lab.

Personally, I'm glad Windows is getting in the game, just like I'm glad when the US gets competition in the space program. Competition, it's a good thing (tm). :)

Re:C# may take over this market

[dvdeug]

C# is a winner for these kinds of applications: it is far simpler and less error prone than C++,

For a community that still uses Fortran? I don't think that's their biggest concern.

Sun really screwed this up with Java: if they had taken the scientific and numerical communities seriously and added the necessary features to Java, Java could be the undisputed winner in this market.

Standard Java requires IEEE floating point, so Java programs run the same everywhere. A community that used Crays (which were renown for their lousy, but fast, floating point) doesn't want their programs to run everywhere with precise but slow mathematics; they want their programs to run on their hardware with the hardware floating point as fast as possible.

ultimately turned into a bloated web applications platform.

Isn't that what Java is? It sounds like you're asking Java to service an audience completely different from what it was designed for.

C# is a winner for these kinds of applications: it is far simpler and less error prone than C++,

For a community that still uses Fortran? Not my biggest concern.

I don't see either Java or C# offering the raw speed that the scientific community wants. Speed and predictability come first, not portability or security (scientific code never needs to run as root and frequently runs on computers disconnected from the internet.)

Re:Slashbotters and FUD

[A beautiful mind]

You can make a turtle fly at speeds breaking the sound barrier aswell.

Is the turtle designed for that?
Is it capable of that?

Please answer those questions, then replace the turtle with windows and breaking the sound barrier with building a cluster. As someone noted, MS and Dell put together that cluster. The obvious conclusion is that other people aren't out of their minds yet.

Also, i said "No wonder that the top500 is mostly made up from unix/linux systems". The exception makes the rule stronger, is this familiar?
According to this article: "the computer scientists behind the Top 500 list say that 291 of the machines (58 percent) on the list are clustered machines. While the Top 500 list does not specifically identify the operating system platform, it probably breaks down to around 55 percent Linux, 40 percent Unix, and another 5 percent as Windows platforms."

Re:Slogan

[susano_otter]

What kind of things do they want to customize? Disk IO? Memory paging?

Most customization is application customization, not OS customization. And customizing an app is identical to writing an app. And people write apps for Windows all the time. A wide range of apps, doing a wide range of things. So you can't be talking about app customization.

But if you're talking about OS customization, could you at least wait until you see what the MS Compute Cluster implementation includes, before you complain?

For all you know, the whole point of releasing a cluster-specific version of the OS is to include more of this flexibility you insist is so important.

Re:Slashbotters and FUD - link with rank please

[Herschel Cohen]

I followed the link, then downward several levels without seeing any claim this was ever in the top 500. Perhaps I just missed the text. But there were pricing and offers of help, nonetheless, the claims seemed to be circumspect given the performance level of your claim.

Is this my failure, or is your link just FUD too?

So...

[bonch]

You can make a turtle fly at speeds breaking the sound barrier aswell.

Is the turtle designed for that?

Yes.

Is it capable of that?

It's in the Top500 list, isn't it? If it wasn't capable, it wouldn't be doing it. Simple as that.

Please answer those questions

Just did.

So, basically, you implied Windows isn't good enough to run as a Top500 server, someone pointed out that it already does, and now you're defending it by saying, "Even though it is, it's still not good enough?"

This kind of crap really makes the community look immature.

Re:So...

[rewt66]

> > You can make a turtle fly at speeds breaking the sound barrier aswell.

> > Is the turtle designed for that?

> Yes.

Too many... sarcastic replies... can't choose...

This has to rank as one of the stupidest statements ever made. Dude, I don't know what things are like on your planet, but around here, turtles were not designed for any such thing. They were designed for swimming in a much thicker medium at much slower speeds.

In the same way, Windows was not designed for clustering. It wasn't even designed to be multi-user or Internet-enabled, and we've seen the security problems that have resulted from Microsoft kludging it to do what it was never designed to do. (And I'm sure you're going to say "Of course Windows 1.0 wasn't designed for that, but NT was!" But NT, while it was (at least supposed to be) a from-the-ground-up rewrite of Windows, it still kept enough of the original design to be seriously flawed with respect to multi-user (see the shatter attack) and the Internet (see the RPC issues, along with many others). Microsoft added to the capabilities, but never fixed the design.)

And, yes, you can make turtles fly supersonic. But the G-force from the JATO does bad things to their internal organs, and the duct tape chafes their hide. In the same way, Microsoft can make Windows cluster. But was it designed for it? Or was it just forced into the role, with a lot of duct tape and bandaids?

Re:Slashbotters and FUD

[Anonymous Coward]

MS donated that cluster and constantly uses it to advertise that WinNT is as good as UNIX/Linux.

Not a very good example.

Wrong solution to wrong problem

[UnknowingFool]

Even Microsoft's Excel can benefit, he said, noting that some businesses have worksheets that can take hours to calculate. Today, such work requires third-party add-ons such as software from Platform Computing. However, Theimer said that Microsoft may be interested in offering that capability itself. "Microsoft is also looking at this," Theimer said.

Am I the only one that read this and think Theimer and MS have no clue what clusters are designed to do? Using a spreadsheet to do lots of number crunching is an application problem. You are using an application in ways that it was not intended. That's far different than an OS performance issue where your application is limited by what the OS can do and need some way to tweak the OS to perform better. How is MS Cluster going to solve this spreadsheet problem? Is that spreadsheet going to run across hundreds of machines? My solution would be not to a use a spreadsheet in the first place. There must be some other more suitable application.

I may be cynical but I see this as another ploy by MS to expand their revenue by bloat. As their software gets more bloated, the average consumer will be forced to use newer and more powerful OS to run it. I'm sure that MS will be tickled in 2010 when you need a 64 way processor just to run BackOffice 2009 so they can charge your company for a license for each of the the 64 processors.