Hi Adrian, I've got diskd configured to be used for objects over 500k - the datacomm run is all 13K objects so essentially it's doing nothing. Interestingly though I see the same stuff if I use ufs only, or just diskd. I am using kqueue - I will try to get you stats on what that shows. If I push it too far (1800 RPS) I can see squid visibly failing - error messages, too much drive load etc. But at 1200RPS it runs fine for > 10 minutes - I'd really like to get this solved as I think there is potential for a lot of performance. I've just run a test now at 300RPS and it failed after 80 minutes -- very weird... I'll try to get you all the stats I can tomorrow morning Thanks again for the help Dave -----Original Message----- From: Adrian Chadd [mailto:adrian@xxxxxxxxxxxxxxx] Sent: Thursday, November 08, 2007 5:37 PM To: Dave Raven Cc: 'Adrian Chadd'; squid-users@xxxxxxxxxxxxxxx Subject: Re: Squid Performance (with Polygraph) On Thu, Nov 08, 2007, Dave Raven wrote: > Hi, > I've been looking for a way to do the profiling, but I'm stuck with > FreeBSD 4 - any ideas? Cache_mem is at 96mb, its almost definitely getting > filled immediately - I've also tried setting it to 8 just to be sure, no > difference... Hm. FreeBSD-4 doesn't have pmc, but pmc is proving to be a bit useless when profiling high-syscall-throughput applications. tsk. > It's a bit difficult to graph -- disk IO I can see with iostat, it seems to > stay the same even after my slow down period... I'll assume you're running with kqueue. I'd run systat -vmstat 1 under freebsd and watch all the key values, see what peaks. Also, are you using diskd when you're not using COSS? Adrian -- - Xenion - http://www.xenion.com.au/ - VPS Hosting - Commercial Squid Support - - $25/pm entry-level VPSes w/ capped bandwidth charges available in WA -
