Re: [Gems-users] Excessive Cache Misses

[Mike Marty <mikem@xxxxxxxxxxx>]

I strongly *not* using MI_example for any simulation.  It was meant as a 
simple code example. 

Are you simulating a 2-processor system?  Is it possible that the OS 
scheduled some other process?
--Mike

Robert William Havlik wrote:
> Hello,
>   I am trying to simulate a microbenchmark which excercises false sharing.  I modeled it off of the multiprocessor parallel microbenchmark.  In my code, one thread writes 32 locations in an array and another processor reads 32 locations in an array.  
>
> I have used Magic Breakpoints after the threads are created and after they end so I can ONLY get the data during that execution of threads.  I also use ruby0.clear-stats to clear data recorded before the threads start.  
>
> However, when I do a dump-results at the end, my cache misses just for user_misses is 66698, while overall total misses is 702887.  Both of these values seem unreasonable.  Where could these be coming from?  Obviously the operating system contributes some cache misses, but still, these numbers seem too high.
>
> I've also noticed that even when I let the system idle for a few minutes (about 2), I get 394000 supervisor misses (0 user misses which makes sense).  I'm currently using the MI_example because I just wanted to look at misses with the simplest protocol possible.
>
> Any insight would be very much appreciated.
>
> Thanks,
> Robert
> _______________________________________________
> 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.
>