Re: [Gems-users] The distribution of "total_misses" in ruby stats output

[Dan Gibson <degibson@xxxxxxxx>]

Which version of Simics are you using, and what is your target machine?

My initial thought is that there are more than one phys_mem* objects, 
and Ruby is only attaching itself to phys_mem0 (in Simics 2.x/Sarek 
target, there is _ONLY_ phys_mem0). However, with some other versions of 
Simics and/or different target machines, there are sometimes phys_mem 
objects for EACH cpu -- verify that you only have one phys_mem* object 
(via the Simics command line), and that its name is phys_mem0.
Regards,
Dan

Lide Duan wrote:
> I found something strange when looking at the ruby stats output files. 
> I am simulating some 16p checkpoints. If I place all the 16p on a 
> single chip (g_PROCS_PER_CHIP 16), the results related to cache misses 
> are shown as follow:
> Total_misses: 2537939
> total_misses: 2537939 [ 2537939 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ]
> user_misses: 1763423 [ 1763423 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ]
> supervisor_misses: 774516 [ 774516 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ]
>
> I suppose the 16 numbers in the brackets correspond to the misses on 
> each processor, but as we can see only misses on the 1st processor 
> were observed. On the other hand, I also got the followings:
> instruction_executed: 1322569080 [ 82094180 83437607 81021513 81093394 
> 82632809 122205588 82083483 80671650 81138678 80191395 79408095 
> 79763896 81353502 81528474 62244475 81700341 ]
> cycles_per_instruction: 4.63392 [ 4.66589 4.59077 4.72767 4.72348 
> 4.63548 3.13441 4.6665 4.74817 4.72084 4.77661 4.82372 4.80221 4.70837 
> 4.69827 6.15384 4.68839 ]
> misses_per_thousand_instructions: 1.91895 [ 30.915 0 0 0 0 0 0 0 0 0 0 
> 0 0 0 0 0 ]
> So definitely all the processors were running something, but why were 
> the misses on the 1st processor observed only?
> To address this problem, I tried different configurations. If I place 
> the processors on 4 chips each containing 4p, the first 4 numbers in 
> the total_misses brackets are not zeros. Also, if one processor on one 
> chip (totally 16 chips), all the 16 numbers are not zeros. Therefore, 
> I guess the numbers indicate the misses on each CHIP, not each 
> processor. Am I right? or did I miss something here? Actually I tried 
> different workloads with different network topologies, but got the 
> similar results. Can anybody give me some explanation?
> Thanks,
> Lide
> ------------------------------------------------------------------------
>
> _______________________________________________
> 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!