[GE users] SMP idea for containing multithreading
reuti at staff.uni-marburg.de
Mon Nov 3 10:02:45 GMT 2008
Am 03.11.2008 um 10:09 schrieb Chris Jewell:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> Hi Joe,
> Many thanks for the reply!
>> R/Parallel was released (recently) and it has MPI extensions.
>> runs in parallel, though there are license issues to deal with.
> Yup, I was aware of R/Parallel, though I think this may be beyond
> the scope of my users ;-)
>>> execution node. So, firstly, is there a standard way of limiting
>>> number of processors that a user's job can use /on the execution
>>> I'm suspecting no...?
>> Yes ... through the OS, and through the scheduler ... but the first
>> isn't as easy, and the second may not be respected by the application
> Could you point me at some further information on this -
> particularly the OS idea? I
> assume the scheduler idea means setting some environment variable
> (eg OMP_NUM_THREADS)
> to the number of slots requested?
>> Regardless of the specific VM technology, what you want to do is
>> possible and people have been doing this. The question is whether or
>> not the impact will be substantial, and this is in part determined by
>> the product, the application mix, and what code paths the application
>> itself takes.
> Ok, thanks for that. I'll have a read further, and maybe try out a
> virtual arrangement
> in the next few weeks.
there are already some papers and scripts for this http://
But I think it's too much for just binding processes to cores. With
the "taskset" command in Linux it should be possible to limit the
execution to certain cores inside the sge_shepherd. The only problem
for now is, that every user can use "taskset" to use more than the
granted cores. But IMO there will be an option to restrict the use of
taskset to root in the kernel some day
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net
More information about the gridengine-users