[Beowulf] Similar to a multi CPU machine?
Robert G. Brown
rgb at phy.duke.edu
Thu Nov 17 04:07:45 PST 2005
On Wed, 16 Nov 2005, Jake Thebault-Spieker wrote:
> So, does the openMosix 2.6 development branch function correctly? I
> understand it's not considered stable as of yet, but for my purposes
> will it work? I supposed I could use 2.4, I'd prefer 2.6 though. Do you
> have any other suggestions? Would running an Xserver still be limited to
> one CPU though? Like if I wanted to run icewm, it would be limited to a
> max of 166Mhz, but firefox inside of icewm would also have ~166Mhz? Is
> that correct?
I can't help you with OM stability as it has been a while since I used
it -- I prefer to use tools that let you ride the current kernel, since
I use mostly new hardware and driver issues are difficult enough as it
is.
However, yes, an "SMP" computer in general runs each application thread
on a single processor unless fairly considerable effort is made to
parallelize it. Effort that has to be made with some attention paid to
the particular interprocessor communications architecture. That is, if
you DID have an application that forked off some threads in what was
presumed to be an SMP machine with common memory, the tuning of the IPCs
compared to CPU (after those IPCs were shipped back over the network
instead of across a memory bus) might well result in an application
running more slowly!
OM is a tool that I think of as being correct for a) independent
single--threaded applications where you just want to run them from a
single spot without worrying about which cluster CPU they run on; and b)
one of several ways to help allocate cluster resources for truly
parallelized applications. However, even on an actual dual or quad
processor system, applications don't parallelize themselves.
Yet. Maybe not ever, but who knows what the future holds?
rgb
--
Robert G. Brown http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567 Fax: 919-660-2525 email:rgb at phy.duke.edu
More information about the Beowulf
mailing list