On Wed, 22 Jun 2005 19:35:37 -0500 Gordon Grieder <[EMAIL PROTECTED]> wrote:
> Hi, > > We've been testing a squid proxy at my workplace (~300 machines > locally) on a smaller group of 60 machines. (used the Windows' > "autodetect proxy" thing with some javascript on a local webserver to > get get our 'volunteers') > > Our new machine arrived which will be replacing this test unit. P4, > 3.4 GHz, 2 SATA drives, 1 GB RAM, Broadcom (bge) gigabit NIC. We > connect via gig fiber to teh intarweb and CA*Net4 (research/education > network much like Internet2 in the US) so the users expect pretty > quick response time. > > Like the test unit, I'll be tweaking until I find the sweet spot of > cache hits and age with performance. I was wondering if anyone had any > Squid on OpenBSD pointers and/or answers to these questions: The only pointer I can think of off the top of my head is to make sure you raise your maxfiles before you compile squid, the configure script checks that and limits squid to that many fds. Squid uses lots of fds obviously. > - is it any faster to use multiple cache partitions on the disks > or should one on each drive be good enough? Remember these are SATA > drives. SCSI could be picked up if it will make a huge difference. One cache dir per drive is good. SCSI does make a very big difference with squid, its doing lots and lots of I/O operations all over the drive. But with only 300 users you shouldn't be too worried about performance. > - diskd sounds like it's a good performance booster and has worked > fairly well on my limited testbed. Any caveats on an installation of > the size I'm looking at? Nope, diskd works fine. For such a small setup, I wouldn't spend much time playing with squid configuration besides letting it use all your RAM. There's tons of stuff you can tweak, but unless you have lots of users, you really don't need to bother. Adam
