Stephen C. Tweedie wrote:
> > > No data should get more than about 30 seconds behind by default. With the
> > > -o sync option on ext2fs mounts it should be virtually instantaneous.
> >
> > -o sync is inappropriate for slow media as it ends up not coalescing
> > writes, doing multiple writes to the same block, and excessively
> > serialising reads and writes.
>
> Exactly the sorts of problem which supermount fixes: that code just does a
> full fsync_dev() of the device once the last writer on the filesystem has
> closed the device.
I suppose simply making sure the device request queue cannot go empty as
long as there are dirty blocks would be equivalent. Now that we have
device request queues :-)
> It even defers completion of the close() of the last
> writer until the sync has completed, so that if you copy anything to the
> device, you can be sure that the media is synced up once the copy
> completes.
Now this is a nice feature.
-- Jamie
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
