In my cache server, every mount is set "noatime". Since the server is
dedicated to the web caching, I don't need the "last time someone accessed a
given file".
BTW, FYI.. On my cache.log startup:
Max Swap size: 132592000 KB
4x33 GB UW-SCSI2 drives. I fills almost my entire RAM. My cache_mem is set
to only 50 MBs, since the object table goes quite large into the memory.
But, why mind? The disk access is fast enough (avg. 9 ms to access cached
objects).
The OS lies in a 9-GB disk drive. Logs rotated and compressed
(access.log and cache.log, don't log store.log) daily. Monthly, FTPed to
another server, backup and remove the old ones from cache machine
[access.log may reach 1 GB/day].
My cache occupies entire hard disks, with no partitions [mkfs.reiser4
(yeah!) -f /dev/sdb, sdc, sdd, sde].
The advantage? Well, If I want to zero the cache swap, I just stop
squid, umount the partition, kick a mkfs and re-mount drives. Elapsed time?
10 seconds, I say.
rm -f /usr/local/squid/var/cachexx ?
Damn, it may take up to 20 minutes (with a 27 GB cache_dir).
Another thing I have into account is the fact of the intense I/O in the
cache dir. Definitely, I don't feel comfy about all this I/O in my OS main
disk. And, placing in different disks, the log writting isn't concomitant
with a eventual disk op cache-related.
A power loss might led to a long fsck (the cache mounts aren't FSCKed),
which results to long time bringing the machine back up and lots of users
whining (altough we use WPAD with quite good results when directing the
traffic to another server in case of failure).
My 2 cents: I would consider seriously creating a separate partition to
the cache (if a second or more disks isn't an option). Both of them with
"noatime" ;-)
Best regards,
Rodrigo.
----- Original Message -----
From: <trainier@xxxxxxxxxx>
To: <squid-users@xxxxxxxxxxxxxxx>
Sent: Tuesday, October 11, 2005 6:19 PM
Subject: Re: Which the best OS for Squid?
First off, there's no possible way my cache would "fill" the '/'
partition. There's a cache size directive in squid that's designed to
limit the amount of disk space usage.
Not to mention the fact that I have a utility script that runs every 15
minutes, which pages me if partitions are >= to 90% their capacity. I
mean, honestly, who would run a 146GB cache?
Second off, it's a performance thing. The fact is, the box and the web
run quite fine. This was a test server that was thrown into production
because it works.
My plans to upgrade the device are set, I'm just trying to find the time
to do them. :-)
Thirdly, can someone PLEASE answer my question about setting "/" to
'noatime', as opposed to avoiding it by telling me how and why what I'm
doing
is stupid?
Once again, are there pitfalls to having '/' set to 'noatime'?
:-)
Tim Rainier
Information Services, Kalsec, INC
trainier@xxxxxxxxxx
"Joost de Heer" <sanguis@xxxxxxxxx>
10/11/2005 05:07 PM
Please respond to
sanguis@xxxxxxxxx
To
trainier@xxxxxxxxxx
cc
squid-users@xxxxxxxxxxxxxxx
Subject
Re: Which the best OS for Squid?
trainier@xxxxxxxxxx said:
What if the squid cache is stored on the "/" partition?
That's a bad idea. Your cache could potentially fill up the root
partition.
Wouldn't that be a hideous mistake to set "/" to 'noatime' ?
Wouldn't it be a hideous mistake to put the cache on the same partition as
"/"?
Joost
