Slashdot Mirror
In The Works: Windows For Supercomputers
Robert Accettura writes "According to ZDNet, Microsoft may be feeling threatened by Linux gaining ground in the High Performance Computing (HPC) arena. As a result, they have formed a HPC group to bring windows to these systems. It makes a mention of how clustered computing may be a target. I guess the only thing better than crashing 1 computer at a time is crashing an entire room full at once."
I guess Bill thinks it's time to slow the worlds fastest computers to a crawl. Apparently they aren't crashing enough, too.
by Mike Buddha -- Someday the mountain might get him, but the law never will.
I hope those guys have good firewalls.
Ydco co
"It looks like you are building a cluster, would you like me to tell you how Microsoft can bring it to it's knees?"
Great- when the cluster gets hijacked by spyware and the like- it can send out 3 millions spam emails a hour as opposed to the 5000 a Dell does now.
Is it just me or does the notion of a GUI on high performance computers sound at bit pointless. I thought the point of HPC was to crunch masses of numbers - not something joe average will want to do any time soon. So what's the point of a pretty (and resource hungry) windows interface?
Because every Node needs a Windowing System in Ring 0.
Free as in mason.
Coincidence? Of course not, this has been a strategy since the days of BASIC. Microsoft copies all the good ideas. Of course, it makes a bad and buggy copy, but, hey, that's what a marketing dept is here for, right?
The right to offend is far more important than the right not to be offended. (Rowan Atkinson)
i think billy & co finally figured out how to get big enough iron for longhorn >D
All we need now is a BSOD joke and I'd swear that everytime I read Slashdot it induces a timewarp back to 1998.
Avantslash - View Slashdot cleanly on your mobile phone.
This action from Microsoft is proof positive that they are taking notice of the recent accomplishments of Linux and are trying to counter them with strides of their own in areas that are not their specialty. If nothing else then this is positive for everyone because not only will Linux continue to improve and develop on its own but now both MS and Linux will develop to compete with one another making the overall user computer using experience better for everyone involved. I know everything MS does is looked down upon by the /. majority but this really should be seen as "a good thing".
Please do not let scientific accuracy interfere with the intended humourous/interesting/insightful value of this comment
The same as ever - whenever Windows is mentioned, lots of wisecracks about crashing is posted. Did you imagine they'd port Win95 or Win3.11 to HPC? Duh. They'll port something like WinXP or W2K3, and guess what - those are quite stable OS'es. Of course you CAN make them unstable, but that goes for PenguinWare as well...
Ah well, I better put on my flamesafe suit - I forgot to criticize Microsoft...
Black holes are where God divided by zero
Why in the world would someone want to run a bloated GUI based operating system on hardwared designed specifically to provide services (servers) to it's customers? Unix is great in this aspect as (at least for the most part) running xdm and serving up a graphical interface was intended primarily for end users requiring execution of applications in multiple windows. Unix servers used to NOT run xdm (or any graphical engine) for the purpose of streamlining and providing efficiency and better utilization of system resources. Windows (even in the current Win2003) is far too large for use in a high performance computing environment. Bill my man ... get a clue ... windows isn't for everything!
I find it naturally that MS tries its luck in the HPC world, but windows surely does not fit the bill.
Thomas S. Iversen
.. codename "domino" ?
Living is a horizontal fall
Most supercomputer users aren't going to want to plonk down literally millions of dollars in software licences to Microsoft - they'd rather be spending this money on either plugging in more hardware or on building and refining their analysis engine.
What could MS conceivably offer that would counter this?
The NT kernel only supports up to 32 or 64 CPUs, IIRC. I think it's because the scheduler has one centralised list of CPUs to dispatch threads to, and it quickly becomes a bottleneck for performance. When you have too many threads to dispatch to too many CPUs, this list is completely locked. The MACH kernel has a thread-list per CPU, and dispatches new threads or moves existing threads in a distributed way, so there's no bottleneck (hence MacOS X's performance on clusters ?). I could be completely wrong here, though, correct me if you know better. So my guess is that MS will have to redo the scheduler of the NT microkernel. I don't know about the VM subsystem...
Maybe we deserve this world ?
Friends don't help friends install M$ junk.
I'd invite you to look at Xbox as an example, and the operating system which that runs. There is no requirement for Windows to include a friendly GUI, animated characters, BSODs or any of these other 'hilarious' /. stalwarts.
Please man -
;)
get your facts straight.
First off, the whole GUI environment didn't originally come from Apple (Lisa, or anything else) - it came from Xerox PARC.
your second statement, is nothing but a very good business strategy. Give the users what they want.
your third statement - is unsupported. Do you really think that they JUST NOW started working on this?
and finally - your last statement - simple rebut: Oh yeah, I've never EVER come across any buggy Macintosh/Unix/Linux,(insert OS name here) etc. code. Bugs are natural - we are human, and make mistakes. But at least they do make efforts at patching/new version more often then not.
Ya know, nothings perfect. But no worries here. Your points are typical.
If I had a penny for everyone that bitched about the problems in the world, or in software and did anything about fixing them, I'd totally be a billionaire.
On the other hand, if I had a dollar for every time I saw someone go after a bug and try and fix it, I'd be near broke.
Thanks for the penny.
Oh no, hang on, it doesn't. Ever. I boot up in the morning, switch between video and photo editing software hundreds of times throughout the day with regular use of MSIE and Eudora as well, and then I shut it down at night without it having crashed once. Every day. For years.
Old versions of Windows crashed a lot. Current versions don't. Fact.
This is part of the reason why Linux isn't gaining mainstream acceptance fast enough. Linux advocates talk about all these imaginary flaws in Windows and people out here in the real world think "well that isn't my experience at all". The effect is to create a distance between regular people and Linux advocates, which in turn pushes the mainstream acceptance of Linux further and further away. Linux needs to be seen as "the other big operating system", not some niche software used by a minority who seem to have a totally different experience of Windows than the rest of us.
Well, I guess it's time for everybody else to abandon this space, because Microsoft has it all covered.
In the November 2003 list....
At 68 - a Windows based system at Cornell from Dell with 640 processors (it originally started out at 320 on the list with 252 processors).
At 128 - a Windows based system in Korea with 400 processors.
So Windows doesn't cluster?
Well I agree with you. I do think it more likely that Microsoft would at the very least turn off the graphical part of Windows, remove it completely, or possibly re-write it from scratch.
What I really don't understand is why it would be necessary or smart to brand such a product as Windows at all. Windows means graphical user interface, and the way it's presented ties quite closely to desktop use. It definitely doesn't mean the remote administration that's likely to be required for an HPC, and trying to remotely administer a Windows box is usually quite clumsy compared with a unix box unless you drop a lot of the traditional Windows UI stuff that's often so tied into its operation.
When I think of Windows, and I don't think I'm alone, one of the first impressions that comes to mind is a relatively klunky monolithic GUI-dependant operating system that spends a lot of time drawing pretty front-end pictures. This almost certainly isn't an accurate picture of what's actually happening all the time and it's not to say that Windows couldn't be adjusted to work in other ways. But it's a first impression.
You can at least argue that the graphical side of things is good for usability on the desktop (even though usability realistically takes a lot more than pretty pictures), but why on earth would Microsoft want to continue that image into an HPC market? Surely they have completely different customers in that market with different goals that likely don't include chewing processor time on pretty pictures for the UI.
To me at least, it'd make much more sense for Microsoft to simply create a new operating system here from scratch (or buy a company or whatever they do), and call it something that's not Windows. It could be Microsoft HPC Server, for instance, and be completely independent from Windows. Microsoft can then claim that their new OS specialises in HPC tasks, and it'll also give them an independent OS product to push in the future if either it or MS Windows collapses.
A lot of people seem to be concentrating on the "windows crashes a lot" idea. That's not quite a fair judgement of windows anymore. The only time I've had problems with Windows 2000 and above is poorly written drivers, or anti-virus software. As long as you choose hardware with proven drivers and don't run anti-virus software (firewall it and run minimal services and no IE) Windows should be very stable.
With that said, I think there's other problems with windows as a supercomputing cluster. The first I can think of is lack of a low-bandwidth interface. Linux you can ssh into and get results, control processes, etc. Windows requires a high bandwidth terminal services. In other words it's harder to control remotely.
Other people have brought up the licensing costs, but I'm sure MS would offer huge deals just to get their foot in the door.
I think the biggest problem is just historical and cultural though. The scientific community has a 30 year history with Unix, is familiar with programming in that environment, and has a lot of legacy code that's written for it. They just aren't going to take to a windows environment easily at all.
AccountKiller
It seems that most people here don't know the following:
There is already a kind of high performance Windows server - it's called Windows 2000 Datacentre, it runs on boxes like the HP superdome mainly for bigassed databases. In general these servers are treated like mainframes - they aren't rebooted - they don't need to be!
You don't need to have direct access to the GUI of a windows box in order to use it. Usually you connect using an RDP client, a la X server.
Even mainframes have a local console and these are offen GUI in nature, it doesn't mean that the machines are slow.
Please stop this mindless microsoft bashing - bash them if they deserve it, but as this product isn't available yet, it seems a bit premature to slag it off.
...I install HPC Windows. We run a few SGIs, our biggest being the SGI Origin 3000. We'll probably shift to either a Linux Beowulf cluster or Apple G5 Xserve cluster in the future, since the type of problems we need to solve don't typically need a single image machine using ccNUMA. I doubt Microsoft will be coming up with anything that will be able to run as a large single image for some time now and by then the competition would have moved forward even more. This is Windows HPC Vaporware so competitors will waste time and divide their resources trying to be Windows HPC compatible on their hardware. They did it with Windows NT in the beginning when they supported MIPS, PowerPC, Alpha.... The best strategy would be to ignore Windows HPC, but I know there is a gullible hardware manufacturer born everyday that will buy into Microsoft's sales pitch.
Sure, there are x86 clusters. But there are also an awful lot of IBM supercomputers using Power chips, HP supercomputers using PA-RISC, heck even Apple clusters using PowerPC, SGI machines, Sun supercomputer nodes, and so on. There are a large number of strange and mysterious chips built explicitly for supercomputing that would never be seen in any other kind of use. There are also a large number o different interconnect technologies.
Since Windows is a closed source operating system, are Microsoft volunteering to port Windows HPC to whatever architecture you happen to come up with? What about the bugs that occur when they write this port? How long is it going to take to get Windows stable on an unusual architecture if only Microsoft can change the source but only you can do the testing?
At least with a custom kernel or Linux you can work on the system yourself until it's up and running, and if you're in the business of installing and running clusters/supercomputer, you can probably afford to pay programmers to write an operating system for nodes in that cluster/supercomputer.
Last I heard, the Windows NT 5.x kernel (2000, XP, 2003) was not even endian-clean any more, let alone portable to RISC or VLIW architectures. Why do you think it's has taken Microsoft so long to port to x86-64 and Itanium?
Or are Microsoft going to "mandate" that we use x86 processors for all our cluster needs in the future?
Not to fan the flames, but get real. I run a homebrew GNU/Linux box (still a 2.4 kernel, I'm lazy) at home, and XP at work. At work I can get almost a week out of a boot before Windows chokes on itself and needs to restart. At home, the local power grid and my lack of a UPS determines how often I restart.
Sure, Win2000 and XP are more stable than 95/98 or the travesty that was ME. So it has "come a long way". But let's not be silly and try to call it as stable as GNU/Linux. One crash a week, hell, even if it were once every six months, still seems pretty unstable to me. If that's an "out of touch" point of view, so be it. An OS shouldn't just decide it's had enough and flake out; I don't care how long it's been running.
Anywho, clustering something even the tiniest bit unstable just seems like a funny idea to me. We've all seen Windows behavior when too much stuff is open or a flaky driver has impaired its ability to operate, things gradually failing, the cursor suddenly trapped in just a portion of the screen, swap thrashing as though it were a sign of the apocolypse... The mental picture of racks and racks full of convulsing, imploding Windows boxen when somebody fires up the wrong version of Quicktime is just priceless.
Actually the cornell "theory" center has or at least had a few reasonably large windows based clusters. I did a postdoc over in the CS department ages ago and ported some code over the the linux side. You can basically ssh into the cluster and standard make works (actually I seem to recall having the switch the "/" to "\". The cluster was something like 4 processor boxes glued together w/ myrinet w/ some sort of queueing system. They also had a slew of 2 processor boxes. My experience w/ them was most of the "crashing" had more to do w/ the myrinet drivers and the MPI implementation (which was a commercial MPI). Once those stabilized it ran as well as a normal linux cluster i.e. you submitted jobs they ran :) I went to a day long "windows HPC" conference back then which was a bit entertaining (btw the clusters were basically free for cornell) People only had good things to say about the cluster, but i think its was a bit opportunistic. One thing that was quite obvious was, if machines are free people will run/port to anything *but* when it came to using your own (or grant) money to buy a machine - even over at cornell - which to be honest had quite a stake in "windows based computing" people would go for a linux based cluster (which had already popped up in quite a few departments at that time)
-bloo
I love this whole idea of Windows on a supercomputer! Just think of how fast a spam drone it would make!
Windows only technical asset is a (relatively) good GUI.
And, as we all know, *ALL* mainframes, supercomputers and servers absolutely must have GUIs!
After all,
Memo at Los Alamos Nuclear Laboratory:
"Please be advised that Deep Blue will be rebooted this afternoon at 5:PM in order to complete the installation of Service Pack 11. All jobs currently running and queued will be lost, even those which have already accumulated several years of processor time. We expect Deep Blue to resume normal operation sometime in early August. Thank you for your cooperation, LANL Informatics Department"
Fire and Meat. Yummy.
More info (for the Google-less) and Links....
h tml
/. previously) (10280/17600)
Top 500 - http://www.top500.org/
Cornell - http://www.tc.cornell.edu/
NEC Earth Simulator -
http://www.es.jamstec.go.jp/esc/eng/ESC/index.
The fastest is the Earth Simulator in Japan (35860/40960 Rmax/Rpeak)
Virginia Tech as the fastest Apple cluster (as mentioned on
Cornell has the fastest Windows cluster (1503/3073)
As for the other questions - Google is your friend and the database on www.top500.org is searchable so should be able to answer anything else.
Especially if you have old or weird hardware, e.g. an Aureal sound card.
I have had entire clusters go down due to OS error.
As a Linux advocate I would appreciate it if we could all just focus on promoting Linux rather than putting down other operating systems. Constant attacks against Windows are completely unnecessary; attacks against Linux from MS are necessary for them because that is SOP for MS, but two evils do not make a good. We don't have to be like them. We don't have to use FUD as a tactic.
I don't even understand why this article has been posted here in the first place. The only possible reactions it could have drawn given the context are fear and loathing. If the audience response is predictably going to be overwhelmingly negative, why post in the first place? Is the posting of this sort of article equivalent to an Orwellian "2 minutes hate" session? Was any constructive discussion expected?
The point is not that its an old list, the point is that there are windows-based cluster solutions out there and have been for a while. The reasons linux is popular with academic clusters are:
- free, duh
- easily customizable and open kernel
- ease of stipping the OS down to minimal levels
Until its free MS won't play in this ballpark in any serious way, although they will probably have PR clusters running here and there.
But, that does not mean that windows can't do it, and can't do it reliably. Windows is my least favorite OS; famiarity breeds contempt. But, if administered with the same care that unix admins administer linux, its just a stable and almost as secure. A poorly administered linux box is just as bad a poorly admininistered windows box, imo.