Hi, On Mon, Jun 03, 2002 at 02:13:05PM +0200, chacron1 wrote: > Hi Stephen, > I would like to know the behavior of ext3 for a write() request used > with > O_SYNC > for 64 K in term of disk access method: > - is there a chance to have only one disk access (instead of 16 x 4 K > corresponding to each mapped page ) or at maximum two in journaled mode > ? Journaled mode is guaranteed to produce more writes: it will write to the journal immediately, _and_ will eventually write to the backing store. For overwrites to already-allocated data, ordered data mode ought to allow for better coalesced writes to the filesystem, as there won't be a forced journal flush in that case (for O_SYNC, we don't bother updating the inode in the journal if all we changed was the mtime timestamp.) Cheers, Stephen
