Re: [squid-users] capacity of squid

On Fri, Mar 16, 2001 at 03:27:48PM -0800, Joel Jaeggli wrote:
> I have reiserfs running on our test box... I'll be moving our main
> cache boxes over once we settle on a kernel combination we're
> comfortable running...
>
> ext2 is actually pretty fast, especially if you start comparing it to
> ufs prior to soft-updates. one of the things we've been looking for
> out of reiserfs is less brittle filesystems, so I've been spending
> some time cutting the power on live filesystems, with relativly little
> carnage thus far...

that's actually a really crucial advantage for reiserfs. i use it on all
my squid boxes because reiserfs has the system back up and running in a
few seconds. e2fsck could take literally hours.

the fsck time for ext2 on a large partition with many thousands of files
meant that it was faster and cheaper (downtime means bad reputation and
lost customers and lost income) to just reformat the partition. e.g. a
power failure that took out my 3 LVS load-balanced proxy servers would
mean throwing away nearly 100GB of cached objects (3 servers x 4 cache
drives x 9GB per drive, running at about 90% full =~ 97GB).

reiserfs is pretty stable too. until the recent replacement of the huge
UPS in the server room, 2 of the proxies and the LVS dispatcher box had
an uptime of 374 days each (the third proxy had an uptime of only about
120 days due to a hard disk failure last year). they've been running
flawlessly for over a year. was sorry to have to reboot them.

this was with kernel 2.2.14 + reiserfs 3.5.17 + some other patches.

i'd recommend reiserfs for any linux based squid box.

craig

--
craig sanders <cas@taz.net.au>
      GnuPG Key: 1024D/CD5626F0 
Key fingerprint: 9674 7EE2 4AC6 F5EF 3C57  52C3 EC32 6810 CD56 26F0