On 14.6.2005, at 12:14, Timo Sirainen wrote:
On Tue, 2005-06-14 at 01:41 -0700, Loren M. Lang wrote:
Linux 2.4 kernel on redhat 9 using reiserfs on all partitions with raid1 on 2 100 gig 7200 rpm wd drives. P4 2.4G 800Mhz FSB w/HT but HT has been recently turned off do to the vulnerability in it. Also 1G DDR 2700 ram. It's pretty close to the example config file so that option is turned off. I could try enabling it, but I am currently running 0.99.x, with it's own default config modified slightly, but it that option is still disabled and runs just fine so I don't think that is the slow part. Some difference between stable and 0.99.x seems to cause the slowdown. I could compare the two config files, but I think it's a difference in implementation rather than a different configuration.
How much slower is 1.0-stable, when moving how many messages and how large the source and destination mailboxes are? I can't really think of any difference that would make 0.99.x faster, so there could be some bug but I'm not seeing it in my tests here.
For everyone: the problem seems to be fsync() and fdatasync() calls which could delay for over a second. I'm not sure if this is reiserfs-specific problem or just specific to this one system. Anyway, these calls aren't exactly required but they try to make it sure that in case the system crashes nothing is left in inconsistent state. I'll see if I could get them called less often, and also disabled optionally..