Danny Thomas wrote:
> is a journaled filesystem like lfs (http://www.hhhh.org/perseant/lfs.html)
> a good choice for squid?
It depends a lot on the filesystem, and I do not know the details of
lfs.
Squid without async-io may benefit from a journaled filesystem if the
journal speeds up metadata updates and provides a background task who
commits the changes to the disk structures.
For a Squid with async-io the journal as such will only slow things
down.
Some filesystems are slowed down a great deal by a journal. Others are
speeded up due to more intelligent update management and less need for
the application to wait for the update to take place.
> I asked Adrian Chadd about this and he said:
> My theory is that it will actually be nice to use if the storage code
> is modified just slightly to support FIFO object allocation
I sort of agree with Adrian there, but squid will quite likely have it's
own journaled object store suitable for RAID5 by then... Supporting FIFO
object allocation isn't a small task if one want's it done in a nice and
compatible manner.
> I'm looking at this as an interim measure till a squid fs appears -)
My iterim measure is to start Squid without cache drives while fsck is
running, and automatically mkfs the partition if fsck fails / needs
attention. Performance isn't a major problem here yet (have about 100%
performance margin, which ought to give us at least 6 to figure out how
to boost performance from the system).
-- Henrik Nordstrom Squid hackerReceived on Tue Jan 25 2000 - 17:58:50 MST
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:50:42 MST