rihad wrote: > Ken Peng wrote: >> >> >> >>> Hi there. >>> >>> I'm planning to build a new dedicated Squid-box, with amd64 and 4 gigs >>> of RAM, with two cache_dir's on two separate harddisks and Squid-3 >> doing >>> application level striping, all servicing around 6k users. Will two >>> recent IDE disks of 7200 rpm suffice, or I'm better off getting two >>> 15000 rpm SCSI disks on a dedicated controller board? >> >> 15000 rpm SCSI is surely much better than 7200 rpm IDE's. > > I couldn't argue with that! Here's what I think: as fetching a cached > copy of some HTTP resource from a 7200 RPM IDE disk is _much_ faster > than doing it over, say, a 10-Gig network link, passing the traffic > through Squid is going to feel faster for an end user, but never slower. > Will 7200 rpm -> 15000 increase in rpm be _that_ important given enough > RAM exists to cache most frequently accessed disk blocks (4 gigs for ~6k > simultaneous surfers)? I doubt that. The biggest overall impact of higher rotational speed is read service time... Obviously the rotational latency component of read service time drops by more than half on the 15k rpm disk. does that matter with your workload? maybe, maybe not...
