What is Digg?131 Comments
- inactive, on 10/11/2007, -7/+60Windows Longhorn Server -- Supercomputer Edition, basic edition starting at only $49,999.99.
- crazybugger, on 10/11/2007, -3/+44Windows has a share?
I am surprised. - aDJsavedmylife, on 10/11/2007, -2/+39Another Spin on the article
Unix like OSes 100%
Windows : 0%
*1% margin of error. - EmileVictor, on 10/11/2007, -1/+34They obviously aren't including zombie computers. I think they can be classed as "supercomputer" clusters as well.
- Philluminati, on 10/11/2007, -0/+26Zombie Clusters Shares:
Linux: 0% BSD: 0% Mac: 0% Windows 100% - nixonrichard, on 10/11/2007, -4/+26the 0.4% were just people looking to get their Windows Experience Index to "10"
- schestowitz, on 10/11/2007, -6/+27It has gone up about 2% since the last time it got sampled (Nov 2006):
Top 500 Supercomputer Sites - November 2006
,----[ Some stats ]
| Operating system Family: Linux
| Count: 376
| ^^^
| Share %: 75.20%
`----
http://top500.org/lists/2006/11
Sad stories for Windows:
,----[ Quote ]
| Microsoft only had a brief moment on the Top 500 supercomputer list before
| the machine was rebooted as a Linux cluster last fall.
`----
http://www.cbronline.com/article_news.asp?guid=B2413430-EEAF-455C-9D85-706DC25D148D - uzusan, on 10/11/2007, -0/+18With an extra license fee of $500 per additional chip.
- Jugalator, on 10/11/2007, -2/+20Well, the OS X and Windows thing is really not surprising at all. They aren't even aimed for supercomputer platforms. They're at best operating systems for large server clusters (and yes, there's a difference). Anyway, what surprised me was still the Linux share, but in comparison to Unix. I honestly thought Unix would have a greater slice than that. Not too shabby for an OS that started out as one for 100% home use and a hobby project!
- sqrt7744, on 10/11/2007, -0/+17you do know that Linux is just a kernel, right? Rest assured that the supercomputer is not running GNOME/KDE Openoffice or whatever. People log in through terminals and execute code, such as galaxy simulations. The I/O is 100% text. Other tools are then used to draw graphs, usually running locally on a terminal computer.
- drag, on 10/11/2007, -1/+16Nobody should be suprised that you had a couple Windows clusters on there. There has always been Windows in the Top500. Why? Because Microsoft pays some University or whatnot to make sure that you have one.
Originally it was done back in the NT 4 days. Poor buggers couldn't barely keep them up long enough to complete a benchmark. One node going down would take out the entire cluster.
Nowadays it's probably a lot better.
One thing to keep in mind about all this is that with most clusters the best thing the OS can do is stay out of the way.
The smaller the OS, the simplier the OS, the better the OS. It's just a sort of platform to operate the hardware. The actual software is all highly custom Fortran applications using message passing libraries. Highly complex, but not requiring much thought at the OS level.
Linux, being very customizable and having very good hardware support (for these sorts of thing) is ideal. It's very easy to deal with. Very easy to boot from network, very easy to get the source code and tweak everything. It's small, it's fast. It's what you need if your a research scientist.
This is what a 'Beowolf' cluster realy is. Commodity hardware, commodity interlink (generally gigabit ethernet), and commodity software. Cheap, fast, simple X 1000+
Mac OS and Windows both need a lot of tweaking to get up to the level of Linux. Plus you have to pay money for them, plus you have to enter into very restrictive agreements to get access to the source code. Plus the only people realy familar with Mac OS or Windows are the Apple and Microsoft people. With Linux the developers are immediately aviable from all sorts of different places.
It's not until relatively recently that people like IBM, HP, and SGI has gotten Linux to the point were it's usefull on more then just clusters. Now with big single image machines, the more classical super parrellel 'supercomputer', previously only big proprietary Unix was able to scale up to the point were it was usefull. Now Linux is becoming more and more common. With Linux 2.6 and it's rapid development it's now able to scale up to several thousand CPUs per machine.
For example SGI has 1024-way Linux machines in production. If your somebody like nasa you will even buy a cluster of those things. They have something like a 512 machine cluster each with a thousand CPUs.
IBM is doing similar things with it's blue gene. And soon Sun will be getting in on the action.
SGI has a machine now that runs 4096 proccessors in a single system image running Linux. That is damn impressive for source code you can download from anonymous FTP.
But remember, unless your dealing with those classical supercomputers, the best the OS can do is stay out of the way. Linux does that well, as does BSD. - geminitojanus, on 10/11/2007, -1/+16Unfortunately, it's not that easy. The OS still has to take care of network management and memory management, and Linux excels at both. Outside of that, it's also open, which allows anyone to fix anything that may make their particular application run slowly and supports esoteric CPU architectures and memory architectures which are often used in supercomputers. Linux is just the best OS for the job.
- dvsbastard, on 10/11/2007, -1/+14A Server is not the same as Supercomputer...
I would be surprised if the supercomputers running Windows are anything other than Microsoft owned and run experimental machines... - microft, on 10/11/2007, -2/+15Windows on a cluster?! Who the hell was playing computer games on a super-computer?! :D
- lieutenantmudd, on 10/11/2007, -1/+13Anyone know how vanilla most Linux installations are on supercomputers? Obviously Linux is popular because you can freely change and optimize it for maximum performance, but how heavily modified is it?
- GMorgan, on 10/11/2007, -0/+12Really. Linux absolute dominance in this area is well documented. The Windows and Mac share is limited entirely to machines that MS and Apple built themselves so they could say that a top supercomputer runs their OS. They don't do anything worthwhile and are basically just marketing tools.
Really I stopped telling people how much Linux rules in this area since I assumed everyone who'd care knows. - aDJsavedmylife, on 10/11/2007, -1/+12Can render over 1 million BSODs per second.
- BrainInAJar, on 10/11/2007, -3/+13a custom modified linux should suffice for a supercomputer.
on machines of that scale the OS's job is to dispatch the userland code & get the hell out of the way - chrisinsocalif, on 10/11/2007, -4/+12Nice to see an AMD chip near the top. Lets hope AMD can keep up producing good CPU's. Competition is good for consumers.
- buddyfarr, on 10/11/2007, -0/+8that would be because they are not doing simple things such as word processing or email. the emulations such as galaxy simulations or nuclear blast simulations can take days to process. in runs that take that long every cycle counts. if it took 4 days for you to open an email you would want a faster server too.
- geneticerror, on 10/11/2007, -0/+8they are trying to meet the recomended specs to play crysis.
- madvulcan, on 10/11/2007, -0/+8One specific example, from machine that ranked the second there (jaguar), the compute nodes run a very stripped down linux kernel (called catamount). Only the very essential system services is there. There is no interrupt. When you request for a node allocation to run your code, you get the whole node given to you. I know this b/c I work on that machine (grad student). The login nodes are quite different however. They provide various service for debugging, code development, and things like xmgrace, even konqueror file manager, are there.
Actually Supernova science (astrophysics) is the largest app running there, not galaxy simulation.
@ Spuy767
I'm sure cost was never the primary concern. You don't worry about cost on the software after spending millions on the hardware. Stability and scalability was. Cray developed on this kernel. - GMorgan, on 10/11/2007, -1/+8Apple and MS.
- l0rdn1k0n, on 10/11/2007, -0/+7No, people use Linux operating systems because they're STABLE. The problem with servers from companies like Apple and Microsoft is they introduce a GUI into the equation. It's supposed to make the server/supercomputer/whatever more easily customized, but in reality a GUI eats up precious CPU cycles and allows room for error that could cause a kernel panic; which on a supercomputer would probably not be good.
- blackjack75, on 10/11/2007, -0/+6You just can't beat cluster solitaire.
- GMorgan, on 10/11/2007, -1/+7Well, you need to run the anti virus.
- buddyfarr, on 10/11/2007, -2/+8do they make OS's to do work if you are doing research on the atlantic? would kinda suck if only the pacific coast scientists could get any work done... :)
/sarcasm - nights0223, on 10/11/2007, -0/+6Wow mal, are you personally offended by Schestowitz? That post doesn't even make sense. If you don't like him, block him and be done with it.
And what's this?
http://www.plex86.org/linux2/Anyone-Considering-Hiring-micoshaft-windope-butttroturfers-Check-USENET-First--1721.html - badtz, on 10/11/2007, -10/+15Apple's OS is surprisingly high for an OS that's only had a few server iterations! Doubtful it'll dethrone Linux/Unix anytime soon, but it's nice to see them beat out Windows Server .... now time for the desktop platform ;)
- gimpbully, on 10/11/2007, -0/+5We've got 2 linux clusters (well, 3, one BGL) that have been in the or are in the top500. They are almost completely vanilla installs. We basically only modify a few network based kernel tunables for super large frames and super-low latency. Other than that, it's litterally a SLES8/9 install.
- GMorgan, on 10/11/2007, -0/+5That's a pretty simple change. You don't even need to patch to do this, altering the interrupt clock frequency is a standard Linux config option. If that is his example it is a bad one.
- inactive, on 10/11/2007, -1/+6Probably more than that because the market size for those kind of clusters are something like 1000-10000.
- GMorgan, on 10/11/2007, -1/+6BSD can handle more millions of connections per second so can be better than Linux in a server setting but this isn't a server, it's a super computer. Linux scales better than anything bar perhaps Solaris (which of these is best at scaling is often questioned though on actual metrics Linux wins hands down).
- buddyfarr, on 10/11/2007, -0/+5usually supercomputer OS installations are heavily modified to run on the exact hardware that you have in the system. if it is customized for that exact hardware then it would run much faster.
- Philluminati, on 10/11/2007, -0/+5Whats your point Sanchez? That these facts are skewed or minipulated? How cares what it means for Microsoft
- inactive, on 10/11/2007, -4/+8Awesome. I dont think people are realizing exactly what a "supercomputer" is. You can't just shoehorn Windows or MacOSX onto a 128 processor system with 600gb of RAM and a massive raid cluster of several hundred terrabytes. You can run several virtual machines, perhaps, but who cares what you're running on a virtual machine -- there's an underlying OS that's pulling its strings.
This is hardly something for mactards to be flooding in and saying stupid ***** like "OLOL AT WNIDOWS! MACOXS PWNZU," though Im sure it'll be happening plentifully. - geminitojanus, on 10/11/2007, -0/+4http://www.arc.vt.edu/arc/index.php for one...
- elev, on 10/11/2007, -1/+5It's pretty cool that anyone with a newer G80 based Nvidia card could have been number 2 on the top 500 in 1997.
http://www.top500.org/list/1997/06/100
http://forums.nvidia.com/index.php?showtopic=28512 - inactive, on 10/11/2007, -0/+4"There's anything on there besides posix variants?
I am surprised."
Fixed. - init100, on 10/11/2007, -0/+4"supercomputer market...if such even exists"
Of course it exists, where did you think all those systems came from? Granted, most of them are fairly off-the-shelf hardware, but still, hardware vendors need to know what you are talking about if they want to be an option for such purchases. In addition, certain hardware parts are more specialized, such as the cluster interconnects (high-speed networks used internally in the cluster for communicating between the cluster nodes). - init100, on 10/11/2007, -0/+4And last, but not least, many versions are free. License costs for non-free systems would become significant when talking about hundreds or thousands of nodes (computers), money that could otherwise be used to purchase more hardware. High-performance computing is all about maximum computing power for the money, and there Linux excels.
- dajuggernaut, on 10/11/2007, -1/+5have you any idea what you are talking about?
- AdamPieniazek, on 10/11/2007, -0/+4@dentarthurdent1
"i may sound like a total noob for saying this, but why would you care about a few wasted cycles for a desktop environment on a freaking supercomputer?"
A few wasted cycles are crucial on a supercomputer. As far as I know, scientists use supercomputers to test wicked complex models and crunch huge amounts of data, so a few wasted cycles means you're losing a lot of computing power. Plus, these supercomputers are very expensive, so a wasted cycle costs a lot more in resources and opportunity cost than on a regular computer. I'm not an expert on the matter but have read a few articles where the scientists who use these supercomputers complain that they still don't have enough processing power!
The below essay talks about the constant struggle to maximize supercomputing to take advantage of every morsel of processing power.
http://books.nap.edu/openbook.php?record_id=1001&page=9 - aDJsavedmylife, on 10/11/2007, -0/+4Do you even know what you're talking about? I can see software wise how running Mac Os would be different to any other Unixlike OSes for most super-computer- purposes. In fact isn't OS X more POSIX compliant than most Linux/BSD variants?
http://www.apple.com/science/profiles/vatech2/ - GMorgan, on 10/11/2007, -0/+4Supercomputers != Servers. A mainframe is closer to a server. A supercomputer is designed to turn as much computing power at one task as possible. Generally you run embarrassing parallel problems like weather prediction on them.
The traits needed for servers are very different to the traits needed for supercomputers. A server needs to efficiently handle a large number of connections, a supercomputer needs to be able to efficiently allocate thousands of CPU's. - lemcoe9, on 10/11/2007, -0/+3Hes rite. Windows Server is on *a lot* of servers. XServ isn't really that present.
- abid786, on 10/11/2007, -1/+4Just because you pirate Windows and get it for free doesn't mean that it actually costs nothing.
- meatmcguffin, on 10/11/2007, -0/+3You have absolutely no idea what makes up OS X do you?
*N O O ***** B* - lemcoe9, on 10/11/2007, -0/+3As am I.
- kevincross, on 10/11/2007, -1/+4its no surprise at all.
-
Show 51 - 100 of 128 discussions
Digg is coming to a city (and computer) near you! Check out all the details on our 
© Digg Inc. 2009
— Content created and posted by Digg users is