[parsec-users] regarding simulation stopping criterion

kishore kumar kishoreguptaos at gmail.com
Tue Jan 31 13:42:21 EST 2012


Hi Sparsh,

It may be an off topic, IPC is not an appropriate metric for measuring
performance of multithreaded programs.
The following paper might be useful:

"System-Level Performance Metrics for Multiprogram Workloads", Micro 2008.


Best,
Kishore Kumar Pusukuri
http://www.cs.ucr.edu/~kishore



2012/1/31 sparsh mittal <sparsh0mittal at gmail.com>

> Hello
>
>
> I wanted to ask this, since this is important issue in parallel/multi-core
> simulations.
>
> In multi-core simulation or a parallel simulation, one could use different
> possible stopping criterion.
> 1. Run till each core/program has completed at least X instructions. If
> some programs finish early, let it go on, but don't collect its stats.
> Disadvantage: some programs finish quickly and some very late.
>
> Since different programs have different IPCs, there is one more method.
>
> 2. Find IPC (or progress-rate) of each program individually (through
> offline simuation) and then run till each core has completed its X(i)
> instructions, where X(i) is proportional to its IPC. For example if PROG1
> progresses slower than PROG2, then X1 = 100M and X2=400M (for example).
>
> 3. etc...
>
> Is someone aware if method 2 is a valid/accepted stopping criterion in the
> research-field? If so, can you give some feedback or reference for this. I
> will be grateful.
>
>
> Thanks and Regards
> Sparsh Mittal
>
>
>
> _______________________________________________
> 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/20120131/3bc006f7/attachment.html>


More information about the parsec-users mailing list