Re: [Gems-users] About Ruby Configuration Parameters

[Dan Gibson <degibson@xxxxxxxx>]

There could be any of a host of problems, considering the run is only 
for 10k instructions. I would venture a guess that the other 7 
processors are spinning on a barrier or condition variable, waiting for 
processor 0.
Regards,
Dan

msahin@xxxxxxxxxxx wrote:
> Dear colleagues,
>
> I want to simulate one chip and 8 core machine and I have used following
> ruby configurations and other ruby default parameter values with
> MSI_MOSI_CMP_directory protocol.
>
> Host machine:
> AMD opteron +2 GB RAM
> GEMS 1.3.1 + simics 2.2.19
>
> Target machine:
> Solaris 9 + 8 ultra sparc III
>
> I have used these scripts to start up my simulation:
>
> istc-disable
> dstc-disable
> instruction-fetch-mode instruction-fetch-trace
> magic-break-enable
> cpu-switch-time 1
> load-module ruby
>
> ruby0.setparam g_NUM_PROCESSORS 8
> ruby0.setparam g_PROCS_PER_CHIP 8
> ruby0.setparam g_NUM_L2_BANKS 8
> ruby0.setparam g_NUM_MEMORIES 1
> ruby0.setparam L2_CACHE_ASSOC 4
> ruby0.setparam L2_CACHE_NUM_SETS_BITS 16
> ruby0.setparam NUMBER_OF_VIRTUAL_NETWORKS 5
>
> ruby0.setparam_str REMOVE_SINGLE_CYCLE_DCACHE_FAST_PATH true
> ruby0.setparam_str g_NETWORK_TOPOLOGY PT_TO_PT
> ruby0.setparam DIRECTORY_LATENCY 2
> ruby0.setparam DIRECTORY_CACHE_LATENCY 1
> ruby0.setparam NULL_LATENCY 0
> ruby0.setparam NETWORK_LINK_LATENCY 4
>
> When I used these parameters, I took following ruby results after 5
> billion simics run to warmup cache and 10000 Ruby+opal run for Spec 2001
> OMP benchmark (swim)
>
> Profiler Stats
> --------------
> Elapsed_time_in_seconds: 2875
> Elapsed_time_in_minutes: 47.9167
> Elapsed_time_in_hours: 0.798611
> Elapsed_time_in_days: 0.0332755
>
> Virtual_time_in_seconds: 105.35
> Virtual_time_in_minutes: 1.75583
> Virtual_time_in_hours:   0.0292639
> Virtual_time_in_days:    0.0292639
>
> Ruby_current_time: 38486
> Ruby_start_time: 1
> Ruby_cycles: 38485
>
> mbytes_resident: 357.395
> mbytes_total: 526.738
> resident_ratio: 0.67852
>
> Total_misses: 868
> total_misses: 868 [ 811 8 13 11 7 6 5 7 ]
> user_misses: 318 [ 318 0 0 0 0 0 0 0 ]
> supervisor_misses: 550 [ 493 8 13 11 7 6 5 7 ]
>
> instruction_executed: 321942 [ 10002 45848 43549 38695 40770 47727 49929
> 45422 ]
> cycles_per_instruction: 0.956321 [ 3.84773 0.839404 0.883717 0.994573
> 0.943954 0.806357 0.770795 0.847277 ]
> misses_per_thousand_instructions: 2.69614 [ 81.0838 0.17449 0.298514
> 0.284274 0.171695 0.125715 0.100142 0.15411 ]
>
> And as you see, user misses only occurs in Cpu0 but i have 8 core and
> benchmark is compiled with OMP support so why other cores dont have any
> user miss.
>
> I am not sure that i have used relevant configuration values and protocol
> for my simulation.
>
> would you suggest me which configuration and protocol I must use for my
> simulation.
> Thanks.
> mustafa sahin
>
>
> _______________________________________________
> 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.
>
>
>