httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Bill Stoddard" <b...@wstoddard.com>
Subject RE: Apache 2 cycle vs instruction profiling
Date Wed, 03 Jul 2002 01:27:42 GMT

> Bill Stoddard wrote:
>
> >Subroutine Name          Source File                  Ratio Enter  %
> >Cycles
> >===============          ===========                  ===== ===== ====
> >======
> >.core_input_filter       core.c                        3.15 11
>   1.5 3628
> >.ap_rgetline_core        protocol.c                    3.05 6
>   1.3 3241
> >
>
> Are these numbers from the latest version of ap_rgetline_core(), with
> the optimization to re-use the same temp brigade for the lookahead read?
> If so, I'll proceed with the next wave of fixes on that function (using
> a persistent brigade) to try to optimize it out of the top 5.

These results are from 1.107 and the optimization went into 1.108. I'll
rerun the profile with HEAD and post the results.

>
> >.get_filter_handle       util_filter.c                 4.58 2
>   1.1 2705
> >.ap_find_token           util.c                        4.16 6
>   0.9 2306
> >
>
> Wow, I've never seen ap_find_token() rank so high in a profile before.
> Does your profiling tool show any information on where it's
> getting called?

ap_set_keepalive calls ap_find_token three times. Ooops, I see I am calling
ap_set_keepalive twice (once to many) in mod_cache. I'll correct this.

>
> >.fix_hostname            vhost.c                       3.65 1
>   0.7 1626
> >
>
> fix_hostname() is spending all its time in string processing and
> string-to-int
> conversions.  There's probably an optimization opportunity there.
>
> >.ap_update_child_status_from_indexes scoreboard.c     17.94 2
>   0.6 1435
> >
>
> Is the high ratio here just a result of the cost of writing to the
> scoreboard
> in shared memory?

I suspect so.

Bill




Mime
View raw message