On Fri, 18 Mar 2005 17:22:51 +0100 (CET), Henrik Nordstrom <hno@xxxxxxxxxxxxxxx> wrote: > On Fri, 18 Mar 2005, Jorgen Rosink wrote: > > Yeah, that's why I have no clue how it could be that sometimes there's > > a stall for about 5 seconds client side (congestion free network btw). > Try the following > squidclient -g 60 /squid-internal-static/icons/anthony-unknown.gif & vmstat 1 60 > > is there any irregularities observed in the output of the above when the > clients stall? Will try that next monday, when there's some load... > One thing which I experienced at a customer recently is if you are using > the Linux ext3 filesystem. It is then quite important that the cache > filesystem is mounted with "noatime", and also a good idea to increase the > ext3 journals size. If not the ext3 filesystem will temporarily "lock up" > while flushing the journal when under high load, effectively halting > Squid. This can be seen in the output of the command above as periods > with a lot of pageout activity with no responses from Squid. I'm using reiserfs (3.6), just for the cache volumes, mounted with notail and noatime options. I'll look for the periods anyway ;-). Thanks, Jorgen Rosink
