> -----Original Message----- > From: Finnur Örn Guðmundsson - Skyggnir [mailto:fog@xxxx] > Sent: Wednesday, February 16, 2005 11:36 PM > To: squid-users@xxxxxxxxxxxxxxx > Subject: [squid-users] High loads on linux box running squid - tuning > > > Hi all, > > Im running squid on HP Proliant DL360 (3.06Ghz Xeon / 1Gb ram / 2x 15K disks in raid1) > > We have around 2000-2500 concurrent clients at daytimes > > We recently upgraded the squid box from 866Mhz PIII / 1Gb of ram. It had 100% cpu loads all the time and it was...slow :) > > After i upgraded the box to the 3Ghz Xeon its still performing pretty much the same (Loads from 0.80 to 2.00). Now i am running Squid 2.5.STABLE3-6.3E2 (from RHEL 3 Update 4). I am by no means a squid expert (go figure:) but here are the parameters i have changed from defaults (after googling for awile). > > cache_mem 128 MB > cache_swap_low 80 > cache_swap_high 100 > maximum_object_size 1024 KB > cache_dir aufs /var/spool/squid 4000 16 256 (i know i do not have alot of cacheing space, but this is a temp proxy solution) > cache_store_log none > request_body_max_size 12 MB > half_closed_clients off > > The system is running 2.4.21-27 in non hyperthreading mode. (1 cpu) > > Is there anything i can do to quickly lower the load? How many users can one put on a box like this? I have been google-ing alot and have seen talk about 2500-5000 on one box, not unlike this one. Is this wrong? > > Kær kveðja / Best regards, > Finnur Ö. Guðmundsson > System Engineer - System Operations > fog@xxxx No one has asked what your ACLs look like. They can have a major impact on performance (especially if you are using any regex rules). 60 requests/sec should be simple for a server of that power to handle. I have a 3GHz Dell box running RH9 (Squid2.5STABLE5) that peaks over 70 req/s, using less than 65% of the CPU. System load has never broken 1.50. This box is also serving requests it receives via satellite, which seems to push CPU usage up dramatically. So, my question is... What do your ACLs look like? Chris