[Bioclusters] Benchmarks on a dual core Xeon

Kathleen kathleen at massivelyparallel.com
Fri Jan 20 13:43:48 EST 2006


Chris:

We find that we can mitigate the binding size problems with our distributed
approach, which allows end-to-end scaling to be such that we can assure
enough processors to guarantee that we don't have indices exceeding RAM.
We'd like to download your test suite and run a benchmark on our system to
demonstrate our scaling efficiencies.

Cheers,

Kathleen Erickson
Massively Parallel Technologies, Inc.

-----Original Message-----
From: Christopher Dwan [mailto:cdwan at bioteam.net] 
Sent: Friday, January 20, 2006 10:58 AM
To: Clustering, compute farming & distributed computing in life science
informatics
Subject: Re: [Bioclusters] Benchmarks on a dual core Xeon


Joe,

Good catch. Sorry to be confusing.

It's pretty clear that on NT, the results are cached in memory.  I think
that I mention that a little further down in the same  
section.   For WGS, a much larger dataset, it was certainly bound by  
I/O.  A clearer way to say it would be "BLAST can be bound by I/O unless
available RAM exceeds the space needed for the indices."

After the test period was complete, we had the chance to crank the memory on
the machine all the way up to 32GB.  At that level, even WGS scaled like NT,
so it's certainly a memory thing.

-Chris

On Jan 20, 2006, at 12:48 PM, Joe Landman wrote:

> Hi Chris:
>
>   Interesting results.  In the report you wrote that your blastn nt 
> results were limited by IO.  Could you clarify this?  That is, on a
> 16 GB ram machine, one should expect that the indices are completely 
> cached upon the first scan through the sequences, as the indices fit 
> into much less than 16 GB ram.  We see this happen on 8 GB ram 
> machines and below.
>
>   Thanks.
>
> Joe
>
> Christopher Dwan wrote:
>> I recently had the opportunity to perform some benchmarks (BLAST,   
>> Clustalw, Clustalw-mpi, and MrBayes) on one of the new quad chip,   
>> dual core Xeon (Paxville) servers from Intel.  This isn't a   
>> comparative study between chips, but rather a look at how batch,   
>> multi-thread, and MPI jobs scale on this machine.
>> The report is linked from our main page:  http://bioteam.net
>> Disclaimer:  I did this work in my corporate guise at Bioteam.     
>> Intel provided the hardware, access to their compilers, and they paid  
>> for the preparation of the report.  They did not, however, exercise  
>> editorial control over the content.
>> -Chris Dwan
>>  The BioTeam _______________________________________________
>> Bioclusters maillist  -  Bioclusters at bioinformatics.org 
>> https://bioinformatics.org/mailman/listinfo/bioclusters
>
> --
> Joseph Landman, Ph.D
> Founder and CEO
> Scalable Informatics LLC,
> email: landman at scalableinformatics.com web  : 
> http://www.scalableinformatics.com
> phone: +1 734 786 8423
> fax  : +1 734 786 8452
> cell : +1 734 612 4615
> _______________________________________________
> Bioclusters maillist  -  Bioclusters at bioinformatics.org 
> https://bioinformatics.org/mailman/listinfo/bioclusters

_______________________________________________
Bioclusters maillist  -  Bioclusters at bioinformatics.org
https://bioinformatics.org/mailman/listinfo/bioclusters






More information about the Bioclusters mailing list