On Fri, 2004-11-26 at 15:37 +0100, Daniel Tiefnig wrote:
> Nigel Metheringham wrote:
> > What filesystem are you using?
>
> It's ext3.
> I think you got it.
> Watching the cache statistics of the RAID controler, I can see that the
> hangs in the logging appear when the write hits drop down to about 20%
> to 30%. It then goes up to about 70% to 90%, and logging/work continues.
>
> Data throughput is not very high at these moments (about 8MByte/s each
> read and write to the controler) but all the small files and empty
> directories may sure cause a lot of head seeks on the underlying disks.
When ext3 was very young, we had long discussions on the idea of
separate stable fast journal devices along with data journaling. This
would prevent an awful lot of mail traffic ever hitting the real disk
because the file would have ceased to exist by the time the system got
to flushing the journal transactions to disk.
I still think this would be a neat disk configuration for mail servers.
What I don't know, and it might well be worth seeing if folks in the
ext3 community have an idea, is whether using data journaling on a disk
device with its own caching system would help.
Nigel.
--
[ Nigel Metheringham Nigel.Metheringham@??? ]
[ - Comments in this message are my own and not ITO opinion/policy - ]
