>
> I have a squid 1.1.18 running on a DEC alphaserver 800 box, with
> DEC unix 4.0b. The cache filesystem is 16GB.
>
> When the cache gets a lot of requests (it serves over 2 million requests
> each day, and at peak reaches nearly 50 requests per second) the response is
> horrible. Pages that can be fetced directly from the webserver
> may take a minute or two from the cache. There is no paging going on, and
> neither raw disk I/O nor network I/O seems to be a problem. Of my 16 dnsserver
> processes the last few are rarely used. CPU load rarely goes above 0.5.
>
Forgive me, i'm not that current with squid at the moment, however I
believe the version of squid you are using is going to block on every read
from disk (writes should be cached and therfore no so bad). This may be
made worse if the error message you quote actually indicates a physical
problem with one of your disks. Is there any logging of disk problem
elsewhere?
To work round the blocking problem, if indeed it is your problem I suggest
either:
Investigate the async IO code in squid 1.2. I believe this was intended for
Solaris so may require some work. Does anyone have the async code running on
anything other than Solaris?
Investigate NetCache, a commercial derivative of Harvest, which does support
async IO. However you have to pay for it :-(.
See http://www.netapp.com for details, they may still have a trial version
to download too.
-- GeoffBReceived on Mon Jan 26 1998 - 08:56:42 MST
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:38:31 MST