Hi,
I am not expert, but have couple questions regarding your results.
1. are you profiling the system behavior? if you want to compare the
application performances, then profile the application (according to
"Performance counter stats for 'system wide' (150 runs)")
2. when talking about scheduling, if more threads than available cores, context
switch may need to be counted too.
3. the results say the performances are similar, so either the scheduling has
little impact on performance or other factors impact more such as large data
sets so high memory usage or frequent lock contentions then most time spent on
locks/synchronizations.
I believe there are a lot of other things to consider in this issue too, and
http://clavis.sourceforge.net/ may give you some help.
Thanks!
Best
Junjie
On Saturday, October 4, 2014 10:54 AM, Hamid Reza Khaleghzadeh
<khaleghza...@gmail.com> wrote:
Hi list,
I am studying multi-core scheduling and using Linux Perf as a tool for
comparing different scheduling techniques. So, I usually run a particular
multi-threaded program with different scheduling and obtain some information
when the application is running such as cycles, bus-cycles, cache-misses and so
on. All tests has been done by a Intel Core 2 Duo + Ubuntu 14.04.
I have encountered with wonderful results. In fact, I ran a parallel program
two times with different scheduling. While cycles, bus-cycles and miss-rate are
in test1 are less than those of test2, elapsed time of test1 is more than that
of test2. I wonder whether you could tell me why this happened? The obtained
results were shown at following:
Test1:
# started on Sat Oct 4 18:19:55 2014
Performance counter stats for 'system wide' (150 runs):
1,398,429,117 cycles ( +- 0.13% ) [28.49%]
139,439,143 bus-cycles
( +- 0.12% ) [28.77%]
1,746,982,529 instructions # 1.25 insns per cycle
( +- 0.09% ) [43.18%]
8,596,204 cache-references
( +- 0.45% ) [43.33%]
407,815 cache-misses # 4.744 % of all cache refs
( +- 1.01% ) [42.79%]
15,690,104 r2F77
( +- 0.14% ) [28.51%]
1,123,667,581 rC07D
( +- 0.21% ) [28.15%]
0.365244759 seconds time elapsed
( +- 0.16% )
Test2:
# started on Sat Oct 4 18:34:37 2014
Performance counter stats for 'system wide' (150 runs):
1,403,443,950 cycles ( +- 0.09% ) [28.24%]
140,040,576 bus-cycles
( +- 0.12% ) [28.70%]
1,752,745,189 instructions # 1.25 insns per cycle
( +- 0.10% ) [43.13%]
8,557,756 cache-references
( +- 0.43% ) [43.35%]
412,481 cache-misses # 4.820 % of all cache refs
( +- 1.16% ) [43.01%]
15,728,789 r2F77
( +- 0.16% ) [28.53%]
1,119,849,348 rC07D
( +- 0.18% ) [28.21%]
0.363744760 seconds time elapsed
( +- 0.14% )
------------------------------------------------------------------------------
Meet PCI DSS 3.0 Compliance Requirements with EventLog Analyzer
Achieve PCI DSS 3.0 Compliant Status with Out-of-the-box PCI DSS Reports
Are you Audit-Ready for PCI DSS 3.0 Compliance? Download White paper
Comply to PCI DSS 3.0 Requirement 10 and 11.5 with EventLog Analyzer
http://pubads.g.doubleclick.net/gampad/clk?id=154622311&iu=/4140/ostg.clktrk
_______________________________________________
perfmon2-devel mailing list
perfmon2-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/perfmon2-devel------------------------------------------------------------------------------
Meet PCI DSS 3.0 Compliance Requirements with EventLog Analyzer
Achieve PCI DSS 3.0 Compliant Status with Out-of-the-box PCI DSS Reports
Are you Audit-Ready for PCI DSS 3.0 Compliance? Download White paper
Comply to PCI DSS 3.0 Requirement 10 and 11.5 with EventLog Analyzer
http://pubads.g.doubleclick.net/gampad/clk?id=154622311&iu=/4140/ostg.clktrk
_______________________________________________
perfmon2-devel mailing list
perfmon2-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/perfmon2-devel
