Am 09.10.2002 um 18:19:44 +0100 schrieb Stephen C. Tweedie folgendes: > > This happens when the journal becomes full (and probably also has lots > > of pending data buffers to flush, for data=ordered) > > The data flush happens on each commit, so it's never deferred until > the journal is full. I'm wondering if there's something wrong which > lets dirty metadata accumulate arbitrarily long between commits, but > the normal writeback does seem to be active still --- we do a > set_buffer_flushtime() as soon as the buffer gets marked for metadata > write, so those buffers shouldn't accumulate forever either. > > I'll need to reproduce this and trace it, I suspect. This whole thing happens under these conditions: * On my laptop I unpack a 2.4.19 kernel tree and then apply - say - the 2.4.20 pre 10 patch. During that I will get a lockup with lots of disk activity. * On our main mailserver we see this: If a lot of mail is to be delivered locally, Postfix usually delivers 3-4 message / s to the local maildirs. But every now and then (when tail -f'ing the log), the whole box "does nothing" at all, loggin and operations continue about 18s later. This happens in high load conditions, with all the load bein I/O load. This box uses 2.4.19-ac4 -- Ralf Hildebrandt (Im Auftrag des Referat V a) Ralf.Hildebrandt@charite.de Charite Campus Mitte Tel. +49 (0)30-450 570-155 Referat V a - Kommunikationsnetze - Fax. +49 (0)30-450 570-916 Why you can't find your system administrators: They're seeing a commodities broker, to arrange direct deposit of their paychecks into buying coffee bean futures. -- Gerald (Jerry) R. Leslie jleslie@dmccorp.com _______________________________________________ Ext3-users@redhat.com https://listman.redhat.com/mailman/listinfo/ext3-users
