Re: Your opinion on a http accelerator

Sorry to keep bringing this back to modperl, but thats my
world and I've got to live in it :)
Does your comment about multiplexing imply that keep-alive needs
to be set on the apache side for efficiency when feeding squid?
Normally keep-alive is disabled for apache+mod_perl (when no squid
is used) because nobody can afford the space to keep larger httpds
lingering for subsequent requests: your server-status soon
fills up with "Ks". Does squid avoid this problem?

Finally, is there any risk in changing READ_AHEAD_GAP to 48k
and recompiling?

thanks
-Justin

On Sun, Jan 16, 2000 at 02:22:59PM +0100, Henrik Nordstrom wrote:
> jb@dslreports.com wrote:
>
> > This misses one point - with a pre-forking httpd, and dynamic content
> > generated with buried interpreters and pre-byte compiled modules,
> > such as php or modperl, an accelerator allows the child httpd to
> > write at full speed then become free for another request.
>
> Apache+mod_perl certainly does not classify as a fast or efficient web
> server. However squid will only read-ahead 16 kb (READ_AHEAD_GAP in
> defines.h), but few replies will be in that size I guess.
>
> pre-forking has not so much to do with it, forking with large per-thread
> data state and not handling new connections while the old one is
> lingering for close has.
>
> Squid also helps by multiplexing several client onto a single persistent
> connection, thereby reducing the number of Apache "threads" required.
>
> > modperl this is absolutely critical given that one httpd can easily
> > exceed 10mb of non-shared memory.
>
> True. Thats why you should preload as much as possible of the modules
> used, and not handle too many connections with a single mod_perl thread.
>
> --
> Henrik Nordstrom
> Squid hacker
Received on Sun Jan 16 2000 - 12:43:03 MST