On Thu, Sep 23, 1999 at 12:54:44AM -0700, Jon Mansey wrote:
> We're building a new Squid box for max speed. It has 6 x 9G Cheetas
> in it plus one for system/logs. It will run FreeBSD 3.3.
>
> My question is, do I use ccd, RAID 0, or something else to make one
> large 50G cache dir, or do I let Squid see the individual disks? I
> have heard that more spindles is good, and the config file lends
> itself to multiple disks being defined.
>
> What would give the best performance?
Because Squid uses non-blocking disk I/O and an event-driven structure,
and does very scattered reads/writes of many small files, I think
you'll get more throughput by leaving the disks independent.
In this configuration you'd be able to have a separate seek for a
cached file outstanding on each of the disks concurrently; with RAID 0
or software striping, there's much more likelihood of a particular
single file retrieval requiring concurrent reads on multiple disks.
That's good if you're doing consecutive reads/writes on single large
files, bad if you're doing many reads and writes of randomly positioned
small files concurrently.
If you're familiar with tuning one other well-known UNIX app, I think
of Squid's access patterns as being "like INN, only more so."
> If any one has any background on this, it would be much appreciated.
> Or if there is a good online resource for Squid hardware
> configuration and optimization?
Not really, though there were some threads on this mailing list a
couple months ago, and some recent discussion of correct RAM to disk
ratios. I think the answer is that nobody's very sure what's optimal,
other than that too little RAM per Gb of disk is bad, and slow DNS is
very bad.
-- Clifton
-- Clifton Royston -- LavaNet Systems Architect -- cliftonr@lava.net "An absolute monarch would be absolutely wise and good. But no man is strong enough to have no interest. Therefore the best king would be Pure Chance. It is Pure Chance that rules the Universe; therefore, and only therefore, life is good." - ACReceived on Thu Sep 23 1999 - 13:52:19 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:48:32 MST