Re: [squid-users] squid performance epoll. 350req/sec 100% cpu

On Wed, 2006-05-24 at 16:37 +0200, Henrik Nordstrom wrote:
> ons 2006-05-24 klockan 14:45 +0100 skrev John Horne:
>
> > Can I jump in and ask if the bootstrap script needs to be run when
> > installing on a Fedora Core 4 linux system using the squid SRPM?
>
>
> If the SRPM spec includes patches which modifies configure.in,
> Makefile.am or any other autotool config file then yes, it needs to
> bootstrap the source tree before compile. Either by running the
> bootstrap.sh script or manually invoking the autotool programs
> separately.
>
In which case, and as far as I can tell, the bootstrap script doesn't
need to be run in this instance.

Also, grepping the squid source, it seems that only the epoll patch
makes use of ENTRY_DEFER_READ. Since this appears in our log files epoll
must be being used.

Having said that though, I have just run strace on the main squid pid
and captured the output to a file. I let it run for a couple of minutes
- we have a busy cache - so the file was getting big. The file contains
no mention of poll/epoll but lots of 'select' (over 30,000).

It is very confusing that our squid seems to have the epoll patch
included, but is using select as well as producing error messages
produced by the epoll patch.

Any ideas as to what is going on here?

John.

-- 
---------------------------------------------------------------
John Horne, University of Plymouth, UK  Tel: +44 (0)1752 233914
E-mail: John.Horne@plymouth.ac.uk       Fax: +44 (0)1752 233839