[Bug: 21.5-b28] performance on aix, cpu hog, another session gprof output

[Thomas Mittelstaedt]

Okay, I have attached the truss output with the start:



Jerry James schrieb:
> Thomas,
>
> Thanks for going to the effort of collecting this profiling data.  It
> gives us some interesting information.
>
> 2008/3/11 Thomas Mittelstaedt <T.Mittelstaedt at cadenas.de>:
>   
>>  ngranularity: Jedes gefundene Beispiel deckt 4 Byte ab. Zeit: 97,87 Sekunden
>>
>>   %   kumulativ   selbst              selbst     gesamt
>>   Zeit  Sekunden   Sekunden    Aufrufe  ms/Aufruf  ms/Aufruf  Name
>>   0,2       0,21     0,21                             .__is_wctype_sb [1]
>>     
>
> [snip]
>
>   
>>   0,0       1,00     0,01                             .xtCreate [48]
>>     
>
> So the top 48 time hogs together consume a grand total of 1.00 seconds
> out of 97.87 seconds of execution.  Your other profile was similar:
> out of 56.55 seconds of execution, the top 43 collectively consumed
> 0.73 seconds of execution.  This indicates that the performance
> problems you are seeing are not occurring in the profiled (i.e.,
> XEmacs) code.  We appear to either be triggering a performance bug in
> some library, or using some library in a suboptimal way.  In either
> case, this is as far as a profiler is going to take us.
>
> I'm still curious about the large number of kioctl calls in the truss
> output you showed us earlier.  Do you know why that output did not
> start with the launching of the XEmacs executable?  Can you get us
> truss output that does start from launch?  I want to know what file
> descriptor is the target of all that manipulation.  It may be a red
> herring, but I think it's worth checking out.
>
> Thank you,
>   
-------------- next part --------------
A non-text attachment was scrubbed...
Name: xemacs-truss.out.gz
Type: application/x-gzip
Size: 113606 bytes
Desc: not available
Url : http://lists.xemacs.org/pipermail/xemacs-beta/attachments/20080311/99d938b5/xemacs-truss.out-0001.bin