Question about Cyrus, ext3, and Linux kernel 2.4.18 bug

[Mika Iisakkila]

Michael Sims wrote:
> Can you recall if you started with the data=journal configuration, or did you 
> switch to it after running under data=ordered?  If so, was it for performance 
> reasons, and did you notice an improvement?

I once tried through all the options on a laptop with a slowish IDE
drive, and the performance difference between data=ordered and
data=journal was something like 1 to 10 while appending messages
to a mailbox. Ordered was rather slow, 2-3 messages per second
on a P3/700. This must be due to Cyrus doing a lot of fsync()s,
which essentially prevents the kernel from doing effective write
cacheing with data=ordered. Don't bother trying data=writeback,
it was abysmally slow, often 1 message per 2 seconds...

I have 2.4.20 + ext3 (from vanilla sources) on all my machines now.
I think I still have data=journal on the laptop because I like to
live dangerously :-) (and it saves a little on disk traffic = battery)
but one apparently file system related crash on another test machine
convinced me to leave them on default options. None of then run Cyrus,
though.

My old message with the results should be found in the list archives
(12th September last year).

--mika