ThreadPoolingMixIn

[pavel.uvarov at gmail.com]

On Jun 3, 1:19 am, miller.pau... at gmail.com wrote:
> On Jun 2, 12:41 pm, pavel.uva... at gmail.com wrote:
>
>
>
> > On Jun 2, 7:15 pm, Michael Ströder <mich... at stroeder.com> wrote:
>
> > Here are benchmarks for FreeBSD 6.2, amd64
>
> > packet_size         x         y
> >           0    499.57   1114.54
> >        1024    499.29   1130.02
> >        3072    500.09   1119.14
> >        7168    498.20   1111.76
> >       15360    499.29   1086.73
> >       31744    500.04   1036.46
> >       64512    499.43    939.60
> >      130048    499.28    737.44
> >      261120    498.04    499.03
> >      523264    307.54    312.04
> >     1047552    173.57    185.32
> >     2096128     93.61     94.39
>
> > x = ThreadingMixIn replies/s
> > y = ThreadPoolingMixIn replies/s
>
> Well, I'd say you've got yourself a winner.  Performance (at least on
> FreeBSD) seems as good or better for your ThreadPoolingMixin than
> ThreadingMixin.  Is this with the default values of min=5 and max=5
> worker threads?

No, I initialized thread pool with min_threads=2, max_threads=200 and
min_spare_threads=20.

For Linux (2.6.22, amd64) I got even more dramatic improvement:

packet_size        x         y
          0   249.97   2014.63
       1024   249.98   1782.83
       3072   240.09   1859.00
       7168   249.98   1900.61
      15360   249.98   1787.30
      31744   238.96   1808.17
      64512   249.85   1561.47
     130048   237.26   1213.26
     261120   249.98    841.96
     523264   249.97    595.40
    1047552   236.40    351.96
    2096128   216.26    218.15

x = ThreadingMixIn replies/s
y = ThreadPoolingMixIn replies/s