Re: [PATCH v1 00/19] Increase resolution of load weights

[Ingo Molnar <mingo@elte.hu>]

* Nikhil Rao <ncrao@google.com> wrote:

> 1. Performance costs
> 
> Ran 50 iterations of Ingo's pipe-test-100k program (100k pipe ping-pongs). 
> See http://thread.gmane.org/gmane.linux.kernel/1129232/focus=1129389 for more 
> info.
> 
> 64-bit build.
> 
>   2.6.39-rc5 (baseline):
> 
>     Performance counter stats for './pipe-test-100k' (50 runs):
> 
>        905,034,914 instructions             #      0.345 IPC     ( +-   0.016% )
>      2,623,924,516 cycles                     ( +-   0.759% )
> 
>         1.518543478  seconds time elapsed   ( +-   0.513% )
> 
>   2.6.39-rc5 + patchset:
> 
>     Performance counter stats for './pipe-test-100k' (50 runs):
> 
>        905,351,545 instructions             #      0.343 IPC     ( +-   0.018% )
>      2,638,939,777 cycles                     ( +-   0.761% )
> 
>         1.509101452  seconds time elapsed   ( +-   0.537% )
> 
> There is a marginal increase in instruction retired, about 0.034%; and marginal
> increase in cycles counted, about 0.57%.

Not sure this increase is statistically significant: both effects are within 
noise and look at elapsed time, it actually went down.

Btw., to best measure context-switching costs you should do something like:

  taskset 1 perf stat --repeat 50 ./pipe-test-100k

to pin both tasks to the same CPU. This reduces noise and makes the numbers 
more relevant: SMP costs do not increase due to your patchset.

So it would be nice to re-run the 64-bit tests with the pipe test bound to a 
single CPU.

> 32-bit build.
> 
>   2.6.39-rc5 (baseline):
> 
>     Performance counter stats for './pipe-test-100k' (50 runs):
> 
>      1,025,151,722 instructions             #      0.238 IPC     ( +-   0.018% )
>      4,303,226,625 cycles                     ( +-   0.524% )
> 
>         2.133056844  seconds time elapsed   ( +-   0.619% )
> 
>   2.6.39-rc5 + patchset:
> 
>     Performance counter stats for './pipe-test-100k' (50 runs):
> 
>      1,070,610,068 instructions             #      0.239 IPC     ( +-   1.369% )
>      4,478,912,974 cycles                     ( +-   1.011% )
> 
>         2.293382242  seconds time elapsed   ( +-   0.144% )
> 
> On 32-bit kernels, instructions retired increases by about 4.4% with this
> patchset. CPU cycles also increases by about 4%.
>
> There is a marginal increase in instruction retired, about 0.034%; and 
> marginal increase in cycles counted, about 0.57%.

These results look more bothersome, a clear increase in both cycles, elapsed 
time, and instructions retired, well beyond measurement noise.

Given that scheduling costs are roughly 30% of that pipe test-case, the cost 
increase to the scheduler is probably around:

	instructions:	+14.5%
	cycles: 	+13.3%

That is rather significant.

Thanks,

	Ingo