Hi, On Fri, Jan 18, 2002 at 02:20:47PM +0000, Zoiah wrote: > > On Fri, Jan 18, 2002 at 01:23:11PM +0000, Zoiah wrote: > > > >> I see in your benchmark that EXT3 is actually performing better than EXT2. > >> How is that possible? Because as far as I know EXT3 is just EXT2 + > >> journalling which means more work for the HD. > > > > It depends very much on the workload. ext3 can often avoid seeks that > > ext2 has to do, because it can flush data out sequentially to the > > journal rather than having to seek to all the bitmap and inode blocks > > when writing out a change to disk. This is especially noticeable with > > some synchronised-IO benchmarks, where ext2 has to seek all over the > > disk for every IO request, whereas ext3 can just append a bit more to > > the journal. > > So, simply put, you can see the journal as a small on-disk cache in those > circumstances because the head doesn't have to search all over the disk? Yes, you can pretty well consider the journal to be a write-back write cache for these purposes. It's a write-back cache with atomic semantics, too, which is of course how the whole recovery-on-reboot works. Cheers, Stephen
