On 22/07/2013 6:08 a.m., Golden Shadow wrote:
> Dear Squid Users,
>
> I found the following squid configure options online, but I did not find enough information about them. What do they mean? Is it better to build my squid with these options?
>
> --disable-poll --enable-epoll , I know this would disable poll and enable epoll, which both wait for some event on a file descriptor, but is it useful to disable poll and enable epoll with squid?
Not really.
>
> --enable-async-io=64
>
>
> --with-aufs-threads=64 , I think this option is implicitly included in the previous one. How many aufs threads should I use? I read it is related to the number of cache directories, would you please explain how it is related?
The above means each AUFS cache_dir is allocated 64 threads for I/O. The
number of threads determines how many parallel operations
(read/write/create/delete) Squid can perform on any one AUFS cache
storage area. It is bounded by the physical disk I/O bandwidth and
controller abilities; too many and the disk I/O controller can't keep
up, too few and Squid gets blocked waiting on threads to become
available. You probably want to have a good read of your disks I/O
specifications to see what the controller can handle before tuning this.
> Is 64 a good value knowing that I configured squid with the following cache_dirs:
> cache_dir aufs /mnt/cachedrive2 1426227 128 512
> cache_dir aufs /mnt/cachedrive3 1426227 128 512
> cache_dir aufs /mnt/cachedrive4 1426227 128 512
> cache_dir aufs /mnt/cachedrive5 427008 128 256
With TB sized drives you would get far more benefit by setting the AUFS
cache_dir min-size parameter and using a rock cache_dir on each drive as
well for small objects.
Squid has a fixed 2^24 count of objects per cache_dir. To fill those
1.4TB caches you need an average object size of 90KB stored there or it
will reach that count limit before filling up.
Amos
Received on Sun Jul 21 2013 - 21:08:24 MDT
This archive was generated by hypermail 2.2.0 : Mon Jul 22 2013 - 12:00:23 MDT