[parsec-users] Benchmark run time in Parsec

Can Hankendi hankendi at gmail.com
Wed Nov 7 17:10:37 EST 2012

It depends on your purpose of benchmarking. 5 minutes of data is generally
enough to capture the power and performance characteristics of the
application (considering you have at least 1Hz sampling frequency). If you
need more data points, you might wanna try putting your machine into lower
p-states, so that the execution time will be longer.

Also note that, you are running applications with just 1 thread (with the
command you provided below). If you run with 12 cores, your execution time
will be much smaller.

If you need to run applications for a longer period of time, you basically
need larger input files, which you can easily create for some applications
(e.g., dedup). But I don't think it would be that easy for most of the
applications, as they might require specific input generation

On Fri, Nov 2, 2012 at 2:36 PM, Greg Shieh <greg.shieh at gmail.com> wrote:

> Hi,
> I am curious about run time of Parsec.
> I have an Intel two sockets system with 6 cores on each socket.
> I compiled ferret with default and ran with -i native.
> I would think it will take 30 mins to complete the testing, however, it
> seems the run time is about 5 mins.
> This may not be good enough for benchmarking. Any suggestions?
> Thanks.
> Greg
> parsecmgmt -a build -p ferret
> parsecmgmt -a run -p ferret -i native
> (3500,3499)
> (3500,3500)
> QUERY TIME: 324.102 seconds
> real    5m24.184s
> user    6m50.439s
> sys    0m1.278s
> [PARSEC] [----------    End of output    ----------]
> [PARSEC] [1] Bienia. Benchmarking Modern Multiprocessors. Ph.D. Thesis,
> 2011.
> [PARSEC] Done.
> _______________________________________________
> parsec-users mailing list
> parsec-users at lists.cs.princeton.edu
> https://lists.cs.princeton.edu/mailman/listinfo/parsec-users
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cs.princeton.edu/pipermail/parsec-users/attachments/20121107/e81bb835/attachment.html>

More information about the parsec-users mailing list