Pthreads performance

[Maxim Sobolev]

Michael Nottebrock wrote:
> On Friday, 11. February 2005 16:06, Maxim Sobolev wrote:
> 
> 
>>Following is result of the run on my 5.3-STABLE/UP.
> 
> 
>>-bash-2.05b$ time ./aqueue_kse -n 10000000
>>pusher started
>>poper started
>>
>>real    0m15.477s
>>user    0m15.432s
>>sys     0m0.009s
>>-bash-2.05b$ time ./aqueue_linuxthreads -n 10000000
>>pusher started
>>poper started
>>
>>real    0m6.118s
>>user    0m2.217s
>>sys     0m0.932s
> 
> 
> Here's what I get on a my UP 5.3-STABLE with ULE and PREEMPTION (thr & lc_r 
> not installed, hence not tested):

Looks very much like either ULE or PREEMPTION pessimize LT without any 
positive effect on KSE. Neither option should have any effect on user 
time, while user times in your tests compare to user times in mine as 
2:1 for both LT and KSE, while real time for KSE is also 2:1 but real 
time for LT is 3:1. Unfortunately I've tested on production machine, so 
that I can't recompile kernel with ULE & PREEMPTION to verify that.

-Maxim

> 
> [lofi at kiste]:~ > time ./aqueue_linuxthreads -n 10000000
> pusher started
> poper started
> 
> real    0m19.157s
> user    0m4.639s
> sys     0m5.545s
> [lofi at kiste]:~ > time ./aqueue_kse -n 10000000
> pusher started
> poper started
> 
> real    0m32.619s
> user    0m31.858s
> sys     0m0.532s
> 
> Still slower, but already lower than factor 2.
>