[Wien] Info please on largest matrix size people have run

[Laurence Marks]

Thanks. Is this with the current version -- just to eliminate some
unintended bug in updates.

In terms of mvapich not working, three ideas (perhaps not useful):
a) My subnet manager (ib) keeps going down and I have to restart it
(/etc/init.d/opensmd start) otherwise I lose IB communications
b) I had to use an updated version of mvapich via svn as the packaged
tar.gz had a bug that broke multithreaded scalapack. Installation was
easy. It may well be that this is not relevant if you are getting it
to work multithreaded on a single node.
c) Put an email on the mvapich list, the group at Ohio State that
develops the package is pretty responsive.

N.B., I rather like mvapich, it is much cleaner than other versions I've used.

On Sun, Sep 14, 2008 at 5:18 AM, Peter Blaha
<pblaha at theochem.tuwien.ac.at> wrote:
> largest matrix size:  75508   (on 64 cpus; i.e. 32 nodes, dual-cpu 3.6
> GHz Xeons + Infiniband; i.e. a rather old (more than 3 years) machine.
>
> ifort 9.0, clustermkl8.0, mvapich-0.9.8
>
> PS: Our sysadmin installed recently ifort10.1, mkl10.0, mvapich-1.0, but
> so far we cannot use mpi across nodes (it just works on a single node).
> One  always gets a stupid mpi.error already at the beginning of the job.
>
>
> Laurence Marks schrieb:
>> I've run into a problem where Iin lapw1  cannot run a matrix of 43695
>> whereas 36867 works fine. These sizes are getting close to the sqrt of
>> the largest integer, so I suspect a problem in either the intel
>> scalapack libraries I am using or the mpi (mvapich).
>>
>> I'm going to try and narrow this down and solve this, but I can only
>> do it for my architecture. It would be useful to know if anyone has
>> run into a similar problem or has run sizes larger than 43695, In both
>> cases please provide information about your architecture, compilers,
>> mpi.
>>
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
>



-- 
Laurence Marks
Department of Materials Science and Engineering
MSE Rm 2036 Cook Hall
2220 N Campus Drive
Northwestern University
Evanston, IL 60208, USA
Tel: (847) 491-3996 Fax: (847) 491-7820
email: L-marks at northwestern dot edu
Web: www.numis.northwestern.edu
Chair, Commission on Electron Crystallography of IUCR
www.numis.northwestern.edu/IUCR_CED