On Mon, Apr 26, 2004 at 01:52:47PM -0500, Steve Lord wrote:
> The very wierd part is the slowdown you see with an external journal
> in the last case. I can see no reason why an external vs an internal
> journal would be different except for the possibility that there
> is some form of interaction with queue unplugging in the elevators
> between devices. It might be worth trying Linus's latest bitkeeper
> tree (2.6.6-rc2) which has per device queue unplugging and see
> if that makes a difference.
I have re-run the benchmarks using the 2.6.6-rc2-mm2 kernel. The
slowdowns still happen, but they aren't as severe.
> I attached a perl script. If you run this once a second and capture the
> output when things are going bad it might tell us a few things.
I saved the output to a directory, one file per second. It is available
at http://untroubled.org/benchmarking/2004-04/xfs_stats.tar.bz2 The
worse slowdowns (200ms average time to list+read+unlink the contents of
a directory) occurred between 14:04:07 and 14:19:12.
Bruce Guenter <bruceg@xxxxx> http://em.ca/~bruceg/ http://untroubled.org/
OpenPGP key: 699980E8 / D0B7 C8DD 365D A395 29DA 2E2A E96F B2DC 6999 80E8
Description: PGP signature