On Sat, Apr 09, 2005 at 07:39:31PM -0700, Peter Gordon wrote: > S. Bergeron wrote: > >YMMV, but I haven't had any problems with ext3 (other than speed issues) > > This is going-off topic, but speed issues? Are you sure you're using > directory > indexing? (This feature keeps the filesystem structure in a hashed binary > tree > rather than series of linked-lists, which improves access time considerably > on > most Ext2/Ext3 filesystems)
Yes, I'm aware of indexing, and I use it on a couple of machines where it's appropriate. It does seem to speed reads quite a bit, although, for overall useage, I've still found it to be slower than JFS on the same hardware (though not as slow as say, FFS on BSD or Solaris). As you posted in that thread, it does help when you're dealing with deep directories. IIRC, it does incur a slight write hit, which on a system that's got rather balanced I/O, might not be worth the tradeoff (e.g. a mail server, or a box running say, Squid). For something where reads are important, however, it speeds things considerably. I haven't found, however, a benchmark that tests things to my satisfaction. For example, I'd like to see a benchmark where they look at directories filled with lots of small files (think Maildir), and also mounted -o sync, testing read and write performance, as well as storage efficiency. I've kind of gotten in the habit of choosing the best tool for the job. On my MythTV machine, I run XFS, because I'm dealing with large files (2.2gb/hr of video). On my desktop it's ext3. On my notebook, I'm still using ext2 for power management reasons (turning off hd with hdparm when running on battery I still haven't gotten to play nicely with any journalling FS). I mostly run BSD on my servers, so I don't have a choice other than slow, reliable FFS. -- S. Bergeron, [EMAIL PROTECTED] -- [email protected] mailing list
