fre 2006-03-17 klockan 15:46 +0100 skrev Matus UHLAR - fantomas:
> many people recommend ext2 over ext3, but note that in case of power failure
> the fsck can take ages.
I took another approach when unsing ext2. I did not have the cache
filesystems in fstab and instead manage them separately from the Squid
init scripts. If the filesystem was dirty on boot I automatically start
Squid without the cache, and have Squid reconfigured once the cache
becomes available. Also automatically mkfs if the standard fsck -p fails
(exit code > 1).
> It might be good idea to put journal on different device for all
> journalling filesystems.
Could be that this would help reducing the spurious lag of ext3. Also
maybe tuning the journal size or commit interval would help. I am not an
expert on ext3 operations/tuning, have only used it in it's default
settings (but with noatime). It is also possible things have improved
since my tests (the tests was run a couple of years ago).
Anyone looking into ext3 tuning should read
http://www.redhat.com/support/wpapers/redhat/ext3/tuning.html
Regards
Henrik
This archive was generated by hypermail pre-2.1.9 : Sat Apr 01 2006 - 12:00:04 MST