Hi, On Tue, Feb 26, 2002 at 11:02:22AM +0100, Heusden van, FJJ (Folkert) wrote: > > Finally, however, we have to do journal replay --- looking for the > > most uptodate copies of each block in the journal and writing them > > back to disk. This is again a sequential scan through the journal, > > but the writebacks are NOT sequential and can cause a lot of seeking. > > I suspect that that is where the bulk of the time is spent. > > What about sorting those writebacks before commiting them to disk? Or > should the disksubsystem take care of that trough that elevator-alg.? The IO layer does that already. We don't want to add any extra intelligence to it in ext3, because that would defeat some of the things the VM is doing internally, such as deciding when there are too many dirty buffers locking up memory so we need to start flushing them out to disk. The ext3 journal recovery just marks all of the writeback blocks as dirty, and lets the VM/VFS schedule the low-level disk writes as it sees fit. Only after the journal scan is complete does the recovery code tell the kernel to force everything to disk. Cheers, Stephen