[openib-general] tests and performance targets

Wed Apr 14 20:52:18 PDT 2004

We will set up an area on the web site for posting test results and
benchmarks.

A framework for automated regression testing of com, laten, allred,
PMB-MPI1, other MPI tests, Lustre, PVFS, IB ULPs, IB access layer, SM,
etc. would be very useful.  Along with needing an open source regression
testing framework, does anyone have an open source testing suite for
IBTA compliance (i.e. IBTA plugfest tests)?  The regression testing
should include tests for performance, IB compliance, stability,
portability, scaling, and manageability.

        - Matt

On Wed, 2004-04-14 at 15:32, Woodruff, Robert J wrote:
> Also pointers to where people can download the benchmarks 
> might also be useful. 
> 
> -----Original Message-----
> From: Fab Tillier [mailto:ftillier at infiniconsys.com] 
> Sent: Wednesday, April 14, 2004 3:18 PM
> To: openib-general at openib.org
> Subject: RE: [openib-general] tests and performance targets
> 
> 
> Could the matrix of test results be posted to the OpenIB site?  It would
> be useful in preventing duplication of efforts.
> 
> - Fab
> 
> > -----Original Message-----
> > From: Jim Harm [mailto:jharm at llnl.gov]
> > Sent: Wednesday, April 14, 2004 3:18 PM
> > To: openib-general at openib.org
> > Subject: [openib-general] tests and performance targets
> > 
> > Here's a partial list of some initial tests we are interested in 
> > automating for evaluating improvements over versions and stacks.
> > 
> > From the Presta1.2 suite:
> > com    measuring throughput and (with options) cross-sectional
> bandwidth
> > laten  measuring latency
> > allred measuring a sequential latency across many nodes
> > and from Pallas:
> > PMB-MPI1       measuring lots of stuff with ping-pong, pong-pong, etc.
> > 
> > We've seen about 820 MB per second with about 5.9 usec latency using 
> > the Voltaire stack on dual Opteron PCI-X nodes.
> > 
> > I'd like to know if anyone can recommend some other public domain 
> > tests and see if anyone has a framework for automation or evaluation 
> > of the test results.
> > 
> > Also, we'll be having to keep track of hardware, version of stack, 
> > version of test, version of compiler, other variables that can give us
> 
> > important information while analyzing the results. If a framework has 
> > been used elsewhere that can be adapted to this application, it would 
> > save time.
> > --
> > }}}===============>>  LLNL
> > James E. Harm (Jim); jharm at llnl.gov
> > System Administrator, ICCD Clusters
> >  Page: 
> > 
> > --
> > To unsubscribe send an email with subject unsubscribe to openib- 
> > general at openib.org. Please contact moderator at openib.org for questions.
> 
> 
> -- 
> To unsubscribe send an email with subject unsubscribe to
> openib-general at openib.org.
> Please contact moderator at openib.org for questions.

-- 
To unsubscribe send an email with subject unsubscribe to openib-general at openib.org.
Please contact moderator at openib.org for questions.