Mailing List Archive

Rob Hartill said:
>
> >From http://home.netscape.com/comprod/netscape_commun.html
>
> -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
> .... The dynamic process-management algorithm
> increases the number of server processes within configurable limits,
> to efficiently handle periods of peak demand. As a result, Netscape
> server software serves documents faster than other HTTP servers, and
> delivers several times greater throughput.
> -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
>
> Hmmmmmmmm.
>

Also not true. "Slightly greater throughput" would be more acurate,
not mention that the greater throughput is eaten up with greater server
overhead.

--
Elizabeth(Beth) Frank
NCSA Server Development Team
efrank@ncsa.uiuc.edu

Elizabeth sez:

Also not true. "Slightly greater throughput" would be more acurate,
not mention that the greater throughput is eaten up with greater server
overhead.

I'm not much taken by discussions of server performance these days --- the
truth seems to be that *any* recent server (Netscape's, yours, ours) gets
tolerable performance on simple file requests for all but the most extreme
loads (approaching the millions of hits daily), barring network and
filesystem difficulties.

Unfortunately, those last conditions (particularly the bit about the file
system) don't seem to be terribly widely understood. For instance, when I
was routinely switching between forking and non-forking Apache versions, I
had occasion to note that the performance hit for setting DocumentRoot to
a path that went through the SunOS automounter absolutely *dwarfs* the
performance gain for pre-forking.

I guess what I'm saying here is that you can only really assess the
performance of a web server in context (including configuration, etc.), and
the server software itself often matters less to the observed performance
than other aspects of the configuration...

rst