httpd-apreq-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Randy Kobes <ra...@theoryx5.uwinnipeg.ca>
Subject Re: [apreq-2] win32 workarounds
Date Sun, 04 May 2003 19:50:19 GMT
On 4 May 2003, Joe Schaefer wrote:

> Randy Kobes <randy@theoryx5.uwinnipeg.ca> writes:
> [...]
> > Failed tests in Performance:
> > 1) perf_init: apr_hash should win: APR = 13.019 vs 
> >    APREQ = 9.013 (microseconds)
> 
> That one is a good (unexpected) failure, since lower
> times are better.

I consistently get that type of result ...
 
> > 2) perf_get_avg: apreq should win (8): APR = 0.000 vs 
> >    APREQ = 1.001 (microseconds)
> > 3) perf_get: apr_table_get(Accept-Language) wins: APR = 0.000 vs
> >    APREQ = 1.001 (microseconds)
> > 
> > Maybe it's a problem with timing things in Win32? I'll
> > take a closer look ... Actually, I'm glad I got these
> > failures, as before they were all passing consistently,
> > and I started to worry that there was something wrong :)
> 
> The results will vary, especially if your CPU is busy
> doing other work.  However, the 0.000 times are suspicious. 
> If you're consistently seeing that, either your machine 
> is considerably faster than mine (Pentium III 500Mhz),
> you're apr compiler (optimization) flags may not be the same 
> as your apreq flags, or maybe there's a timing bug in 
> apr_time_now() on Win32.

The machine is relatively idle, and it's 400MHz, so is
slower. You were right about the optimization flags being
different - changing apreq to agree with apr results in

24 tests run:  22 passed, 2 failed, 0 not implemented.

Failed tests in Performance:
1) perf_init: apr_hash should win: APR = 14.020 vs 
   APREQ = 10.014 (microseconds)
2) perf_get_avg: apreq should win (8): APR = 1.001 vs 
   APREQ = 1.002 (microseconds)

or, shortly later,

24 tests run:  22 passed, 2 failed, 0 not implemented.

Failed tests in Performance:
1) perf_init: apr_hash should win: APR = 15.022 vs 
   APREQ = 10.014 (microseconds)
2) perf_get_avg: apr_tables should win (4): APR = 1.002 vs 
  APREQ = 1.001 (microseconds)

And occasionally all 3 performance tests fail, as before. 
So, like you said, this seems like fine tuning, and we can 
return to it later ...

-- 
best regards,
randy


Mime
View raw message