[Beowulf] Benchmarking?

[Philippe Blaise]

Yes, it's going in the good direction; have a look to the spec hpc / omp 
benchmarks too
http://www.spec.org/hpg/index.html#future

A significant performance benchmark should look like a mean value or a 
collection of mean values
extracted from processor, memory, network, I/O, and throughput measurements
and/or parallel benchmarks of interest. That's why the top500 list is 
meaningless ; please, just read this :
http://www.computerworld.com/governmenttopics/government/policy/story/0,10801,94607,00.html

Au revoir,

 Philippe

Benjamin W. Mayer wrote:

>If you want something that takes into account interconnect more:
>
>http://icl.cs.utk.edu/hpcc/
>
>Cray of course won. But this should give a better reflection of total
>system performance then linpack. Or for the best benchmark of all one
>could benchmark the application(s) they want to run, too bad that is not
>always an option.
>
>Ben
>
>On Thu, 29 Jul 2004, Philippe Blaise wrote:
>
>  
>
>>daniel.kidger at quadrics.com wrote:
>>
>>    
>>
>>>>-----Original Message-----
>>>>From: Jack C [mailto:jack at crepinc.com]
>>>>Sent: 28 July 2004 19:58
>>>>To: Jack C
>>>>Cc: beowulf at beowulf.org
>>>>Subject: Re: [Beowulf] Benchmarking?
>>>>
>>>>
>>>>I was just told about linpack, and that it was good. And
>>>>apperantly there is
>>>>an MPI implementation thereof, but I sure can't find it...
>>>>the MPI standard
>>>>site is not very informative. :(
>>>>
>>>>Anyone know where to find that, or any other good ones?
>>>>
>>>>
>>>>        
>>>>
>>>sigh.
>>>
>>>Linpack is ubiquitous - just look for 'xhpl' in Google.
>>>It will take you to http://www.netlib.org/benchmark/hpl/
>>>Unfortunately this benchmark does not test all aspects of a high performance cluster. Most notably it doesn't really stress MPI, such that most high performance interconnect solutions (IBA, Myrinet, Quadrics et al. [*])  would give fairly similar performance figures.
>>>
>>>
>>>Yours,
>>>Daniel.
>>>
>>>ps. [*] yes I do work for one of those vendors
>>>
>>>--------------------------------------------------------------
>>>Dr. Dan Kidger, Quadrics Ltd.      daniel.kidger at quadrics.com
>>>One Bridewell St., Bristol, BS1 2AA, UK         0117 915 5505
>>>----------------------- www.quadrics.com -------------------->
>>>
>>>_______________________________________________
>>>Beowulf mailing list, Beowulf at beowulf.org
>>>To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
>>>
>>>
>>>
>>>      
>>>
>>Everyone knows that linpack perf is more or less equal to the number of
>>cpus x gemm perf,
>>that's why cheap linux clusters are compared to Quadrics clusters, or
>>even X1, SX6 machines !
>>But what could we suggest to definitively replace this old and stupid
>>and political benchmark ?
>>may be a boycott ?
>>
>>  Philippe Blaise.
>>
>>_______________________________________________
>>Beowulf mailing list, Beowulf at beowulf.org
>>To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
>>
>>    
>>
>
>  
>