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."
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 ?
Athletic Scholarships to universities make as much sense as academic scholarships to sports teams.
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.
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?
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
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.
Sounds like a hardware issue.
In my office we deal with several hundred machines running XP Pro with SP1 and as many patches as exist, and not once have we had this problem of "spontaneous registry corruption" that Linux users always seem to encounter when they run XP for some reason.
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.
I have. And I am not a Linux user. Right now I am using a Mac but my previous computer was a PC running XP. XP wasn't all that bad for personal use... but stability? I had my registry corrupted twice, and it's not like I did anything weird with it.
Maybe it was just me and the linux users... but hey it's not like I was trying my best to make my PC with XP crash or anything like that.
diegoT
You can run Windows without the GUI. (WARNING: this will make Windows fairly useless) Find the key "HKLM\SYSTEM\CurrentControlSet\Control\Session Manager\SubSystems\Required" This lists the subsystems that are started automatically. Remove 'Windows' from the list and delete the 'kmode' key. Now, upon restart, the win32 subsystem won't be started; the computer will stall because it doesn't have anything to do. (winlogon may crash because the GINA depends on win32)
The main problem with running without win32 is that there are (almost) no applications that can interface directly to the native system call interface (ntdll.dll) without using win32. This includes most services.
Some practical examples of Windows without win32 include:
The second part of the first phase of setup, the text mode part in 50 line VGA mode where you partition disks, the full kernel with all the bus drivers are running, but with no win32.
The recovery console.