Re: [Gems-users] instruction count question

[Dan Gibson <degibson@xxxxxxxx>]

Introducing small, random variations in memory latency is known to cause multithreaded benchmarks to take different execution paths, because timing affects the order of memory races and lock acquisitions. Removing the randomness is not a good idea, because by observing only one execution, you study only one possible 'correct' case, when that case may or may not be representative of the true mean.

Regards,
Dan

On Tue, Apr 27, 2010 at 1:30 AM, Jerry Lin <cosjerry@xxxxxxxxx> wrote:

Thanks for reply.

I have some questions about this paper.

It said that this paper introduce small perturbations in memory sysytem timing to evaluate space variability.

I think that it implemented in the ruby/config/RubyConfig.h (memoryResponseLatency()).

If I remove the random() part, then will I get the deterministic simulated result?

But this result might draw an incorrect conclusion when comparing with other results?
And without the small perturbation in memory system, will the time variability still occur when simulating benchmarks?

 

Best regards!

 

2010/4/26 Dan Gibson <degibson@xxxxxxxx>

This is expected behavior. For more information, see:
Addressing Workload Variability in Architectural Simulations,
Alaa R. Alameldeen and David A. Wood,
IEEE Micro Special Issue: Micro's Top Picks from Microarchitecture Conferences, November-December 2003.
Paper: pdf
Original HPCA03 Paper: pdf

Regards,
Dan

On Mon, Apr 26, 2010 at 1:07 AM, Jerry Lin <cosjerry@xxxxxxxxx> wrote:

Hello all,

I have problems when running SIMICS with RUBY.

I ran a benchmark many times using MOSI_SMP_bcast_1level protocol.

But I get very different instruction count between each results.

The instruction_executed varied much more than 5%.

Is there any setting I might miss?

 

Thanks for helping.

 

_______________________________________________
Gems-users mailing list
Gems-users@xxxxxxxxxxx
https://lists.cs.wisc.edu/mailman/listinfo/gems-users
Use Google to search the GEMS Users mailing list by adding "site:https://lists.cs.wisc.edu/archive/gems-users/" to your search.

--
http://www.cs.wisc.edu/~gibson [esc]:wq!

_______________________________________________
Gems-users mailing list
Gems-users@xxxxxxxxxxx
https://lists.cs.wisc.edu/mailman/listinfo/gems-users
Use Google to search the GEMS Users mailing list by adding "site:https://lists.cs.wisc.edu/archive/gems-users/" to your search.

_______________________________________________
Gems-users mailing list
Gems-users@xxxxxxxxxxx
https://lists.cs.wisc.edu/mailman/listinfo/gems-users
Use Google to search the GEMS Users mailing list by adding "site:https://lists.cs.wisc.edu/archive/gems-users/" to your search.

-- 
http://www.cs.wisc.edu/~gibson [esc]:wq!