On Wed 19 Mar, 1997, John Saunders <john@nlc.net.au> wrote:
>Is anybody running their Squid cache_dir on a journaled (or log) filesystem?
>Apparently this type of filesystem has a better sustained write speed since
>all writes are appended sequentially to the journal (log) rather than
Some log filesystems (we're using XFS/SGI Irix) only put meta data
through the log.
Solaris 2+SDS with a translog will put file data through it too unless
you ask it not to (a patch, I believe, and one people have recommended
applying): because you end up sending the real data to the disk,
reading it back and writing it out again. I also believe that if you
have a Presto card, the translog doesn't give you anything extra.
>randomly scattered over the disk. Does anybody know if checkpointing the
>filesystem causes a noticable delay. One day Linux will have one of these
>so I guess I'll be able to experiement then.
It can make directory creation extremely quick: we made the 16*256
directories for a Squid 1.1.8 test in around 10 seconds today. I've
not observed any delays but I'm not 100% up on how XFS works, I've
only skimmed through papers.
-- jrg.
Received on Wed Mar 19 1997 - 06:37:37 MST
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:34:43 MST