RE: [squid-users] Hardware Recommendations

Good point about the multiple cache_dir lines. This is even in the Squid
FAQ, I think... so it doe make sense to not bother with RAID for the cache
dir. Though RAID 10 (if you can get 4 disks) might make some sense for
writing logs (perhaps overkill, though?). Going the JBOD route might work
out well... if rack density is a concern, there are options for dense
external SCSI JBOD chassis out there in 1U and 2U config to add on to a 1U
server (from 4 to 12 external drives); this might be worth your while if you
absolutely need a storage upgrade (in the future, not now, I guess).

Sean

-----Original Message-----
From: Brian [mailto:hiryuu@envisiongames.net]
Sent: Saturday, September 21, 2002 6:05 PM
To: alex@short.net
Cc: squid-users@squid-cache.org
Subject: Re: [squid-users] Hardware Recommendations

On Saturday 21 September 2002 07:54 pm, you wrote:
> I'm only familiar with hardware to scale for accelerator use, but here
> are a few tips:
>
> 1 - If you plan to log with mime_hdrs on (you likely might), dedicate
> more space for logs; I would switch the amounts you have for cache and
> logs.
> 2 - RAID 1 has high read speed, but write speed is both similar to
> a single spindle and dependant upon the caching characteristics of your
> RAID controller. I would be concerned about having your logs write to
> another spindle set than your cache; I would also suggest RAID0 for
> cache, because if your raid gets hosed, you are not losing critical
> data, and you will get better write performance (and more space for the
> price). I don't think that SCSI is key (not from the performance angle
> anyway); suitable IDE raid options are out there too.

If you're considering RAID0, you might as well leave them as separate
filesystems and use multiple cache_dir lines.

That way if one drive goes, you only lose the cache on that drive. The
faster writes are really only on unitasking systems. Once you have a
bunch of concurrent users, the load will balance itself across the two
drives.

Running something like a 3ware in JBOD mode would be a good alternative to
SCSI, although not practical in a 1u case.

> 3 - If you are running redirectors, go the dual CPU route.
> 4 - Opt for more ram if you can.
> 5 - Dual AMD is a better value proposition that dual Xeon any day, but
> cooling must be done right.

Good advice if you rely heavily on redirectors. If it's just the squid
process, the lock handling required in SMP will kill performance.

A P3 with a gig of RAM should suffice.

Like Sean, though, I deal in accelerators, so I can't really estimate what
kind of load 200 workstations will produce.

        -- Brian
Received on Sat Sep 21 2002 - 19:43:05 MDT