[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

RE: EXT3 and disk write back cache


On Fri, 2003-02-21 at 06:29, Robert Liu wrote:

> We have developed a Linux software RAID5 technology, called SR5, which
> guarantees that block writes to the disk array are ordered even if disk
> write back cache is enabled.

That only addresses one part of the problem.

Preserving disk write ordering will guarantee that the data is not
corrupt after a reboot.  However, it will still break applications that
need to know when their data is hard on disk.

Consider things like sendmail(8).  Incoming mail is ack'ed only when it
has been fsync()ed to disk.  At that point, the sending spool is told it
can delete the old copy of the email.  If you crash at this point,
that's OK --- the fsync has guaranteed that the new copy of the email is

If you enable write-behind transparently, then on a power-cycle at this
point, you risk losing the email.

The same applies to all sorts of applications where fsync or O_SYNC is
used to synchronise data persistence between hosts.  Content management
systems, databases, mail and news spoolers and so on all rely on this.


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]