in reply to Decreasing PerlMonks Server Load

To paraphrase Sherlock Holmes,
It is a cardinal sin to optimize before one has data.
Trimming back URLs would indeed same 8 bytes here and there, but does onyone (besides possibly vroom or someone on the Everything Engine team) have measurements that suggest that network traffice is anywhere near a bottleneck? If so, an optimization focused on byte-level traffic might be warranted (though a scheme based on compressing redundant whitespace out of HTML would have a much bigger payoff).

I rather suspect, based on a bit of experience with a similar engine, that the server is CPU bound.

A well-intentioned thought, but one that needs to be pursued after measurement.

Replies are listed 'Best First'.
Re: Re: Decreasing PerlMonks Server Load
by chromatic (Archbishop) on Mar 05, 2001 at 21:56 UTC
    If it's bound by anything, it's bound by memory and the database. I would expect the latter to be the bottleneck, if anything. Building an average page requires several trips to the database.

    I don't know how aggressively the node cache works in the version of Everything running this site. The performance tweaks in the latest version of the Engine (pre 1.0) involve aggressive caching to avoid extra trips to the database and compiling dynamic content.

    It's possible to tune things better, too, by serving images from a different server, by pre-requiring all modules before Apache forks off any kids, and by decreasing the memory requirement of each new kid.

    But sharing a pipe with E2 probably slows things down too.

[reply]

In Section Perl Monks Discussion