[Smeagol-discuss] scaling issues

[Ivan Rungger]

Hello Sahu,

    here are some timing data. I ran the code for 32, 64, 128 
processors, and part of the test on 256 processors. I have no data for 
fewer processors. The figure contains data for runs on two different 
clusters: cluster 1 is an older AMD machine with gigabit ethernet 
connection between the nodes, the other is a newer intel based machine 
with infiniband connections (the newer machine is faster than the old 
one in serial by about 2.3 times). In the figure there are data for two 
systems. The first one, labelled as Fe-MgO-Fe, has 224 orbitals in the 
unit cell (24 atoms), and the second, labeled as Au-Mn12-Au, has 2672 
orbitals in the unit cell (620 atoms).

The speedup as plotted in the lower figures is defined as:
speedup= 32 * t_32 / t,
where t_32 is the time needed for 32 processors, and t is the time. So 
by definition the speedup for 32 processors is 32 (this sets the origin 
of the scale).

Cheers,

 Ivan

Bhagwan Sahu wrote:
> Dear SMEAGOLERS,
>
> If any of you have tested the current version of the SMEAGOL code for scaling i.e, total wall-clock time versus the number of processors on a given computing architecture/s for a given system size (containing say N number of atoms) and have a table/plot of this behavior, I would appreciate receiving it. 
>
> The computing architecture can be cluster of computing cores (with Intel compilers, PGI compilers) as is the trend today or IBM or any other machine.
>
> Thank you in advance.
>
> Sahu 
>
>
>
>
>       Get your preferred Email name!
> Now you can @ymail.com and @rocketmail.com. 
> http://mail.promotions.yahoo.com/newdomains/aa/
> _______________________________________________
> Smeagol-discuss mailing list
> Smeagol-discuss at lists.tchpc.tcd.ie
> http://lists.tchpc.tcd.ie/listinfo/smeagol-discuss
>
>   

-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: smeagol_timings.agr
Url: http://lists.tchpc.tcd.ie/pipermail/smeagol-discuss/attachments/20090515/caf2eb67/attachment-0001.ksh 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smeagol_timings.eps
Type: image/x-eps
Size: 19260 bytes
Desc: not available
Url : http://lists.tchpc.tcd.ie/pipermail/smeagol-discuss/attachments/20090515/caf2eb67/attachment-0001.bin