GEMs keeps track of anything running on each CPU, it doesn't matter if it's an actual application, or just a spin loop. I have more advanced instrumentation in my code to make sure I only count the execution of the applications I want, and not that of the OS.
When you run 8 threads on 16 cores, 8 cores should be running the idle loop. Those will likely have better performance than the other cores. Remember, just because a core is executing instructions doesn't mean it's actually accomplishing work. Spin loops and much OS code aren't actual work.
Phil On Apr 12, 2011, at 12:38 AM, rajath hegde wrote: Hi,
I have a question on the CPU utilization metrics that we obtain in the stat-dump when I run the simulations. I am simulating a 16 core machine with Solaris. When I run a benchmark program with 8 threads, theoretically only 8 cores will be active and it should be reflected in the statistics. But what I am seeing is that the percentage utilization of all the cores are turning out to be the same and I am not able to understand why it is so.
The way I am calculating the percentage utilization is: Instruction_executed * cycles_per_instruction / simics_cycles_executed
Please tell me if I am calculating this the right way or if there is any other way to obtain the accurate CPU utilization? Thanks for the help.
Profiler Stats -------------- Elapsed_time_in_seconds: 12268 Elapsed_time_in_minutes: 204.467 Elapsed_time_in_hours: 3.40778 Elapsed_time_in_days: 0.141991
Virtual_time_in_seconds: 12227.7 Virtual_time_in_minutes: 203.796
Virtual_time_in_hours: 3.39659 Virtual_time_in_days: 3.39659
Ruby_current_time: 45747823 Ruby_start_time: 1 Ruby_cycles: 45747822
mbytes_resident: 1179.05 mbytes_total: 1570.06 resident_ratio: 0.750955
Total_misses: 47008 total_misses: 47008 [ 11925 1232 135 12440 1457 1596 43 1116 1090 422 24 1120 1862 11037 1453 56 ] user_misses: 19092 [ 8509 0 0 9068 101 0 0 36 85 0 0 22 73 958 240 0 ] supervisor_misses: 27916 [ 3416 1232 135 3372 1356 1596 43 1080 1005 422 24 1098 1789 10079 1213 56 ]
instruction_executed: 2848930210 [ 164011837 181023094 182570679 164516946 179675018 180617475 182529107 180468207 180598656 182349335 182613918 180545550 179665794 164810397 180126017 182808180 ] simics_cycles_executed: 2927873679 [ 182991293 182991293 182991293 182992293 182992293 182992293 182992293 182992292 182992292 182992292 182992292 182992292 182992292 182992292 182992292 182992292 ]
cycles_per_instruction: 0.256926 [ 0.27893 0.252718 0.250576 0.278074 0.254614 0.253286 0.250633 0.253495 0.253312 0.25088 0.250517 0.253387 0.254627 0.277578 0.253977 0.25025 ] misses_per_thousand_instructions: 0.0165002 [ 0.0727082 0.00680576 0.00073944 0.0756153 0.00810909 0.00883635 0.000235579 0.00618391 0.00603548 0.00231424 0.000131425 0.00620342 0.0103637 0.0669679 0.00806657 0.000306332 ]
-- Thanks, Rajath Hegde Washington State University
_______________________________________________ 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.
|