Question about custers

Srihari Angaluri srihari at mpi-softtech.com
Fri Feb 7 09:34:07 PST 2003


KNT, I am sure people have constructed mathematical models for computers
(this is a specialization in many computer science programs), but there
are actually benchmark tools available to realistically measure various
performance related aspects of clusters and parallel architectures. See
for example the NAS parallel benchmarks, SPEC benchmarks, and LINPACK
benchmarks. I haven't come across a simple mathematical formula to
directly measure the "power" (which is quite abstract actually) of a
cluster. Although, you can possibly quantitatively measure aspects like
the aggregate network bandwidth, or the total FLOPs of your cluster by
using some formula, or a simple program (see for example Ian Foster's
book on Designing and Building Parallel Programs to get an idea of
performance calculations), the above benchmarks will more accurately
measure these aspects of your cluster and give you a better picture.
Most of these benchmarks are freely available (in binary format as
well). I am sure you will find useful hints for performance calculations
if you look in their documentation. And, fyi, people use these
benchmarks to measure the performance of their clusters to get to the
Top 500 list.

Hope that helps.

> -----Original Message-----
> From: beowulf-admin at beowulf.org 
> [mailto:beowulf-admin at beowulf.org] On Behalf Of KNT
> Sent: Friday, February 07, 2003 11:52 AM
> To: beowulf at beowulf.org
> Subject: Question about custers
> 
> 
> 	Greetz!
> I wanted to ask if there's a way of theoreticaly calculating a cluster
> power by a mathematical formula, basing on the nodes procesor 
> type, ram,
> etc.? Assuming also that the components of each node can be different.
> 
> 						-Thanks from the above
> 							  KNT
> -- 
> ------------------<<KNT>>------------------
> ~ zajcewl at poczta.wp.pl ~ knt at klub.chip.pl ~
> ~                      ~ knt at pf.pl        ~
> ~ GG:3418267           ~ ICQ: 99730430    ~
> ~ Registered Linux User: 300900           ~
> ~ Registered Linux Machine Number: 186168 ~
> -------------------------------------------
> 
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) 
> visit http://www.beowulf.org/mailman/listinfo/beowulf
> 




More information about the Beowulf mailing list