Re: Agenda for ext3 developer interlock call (June 21th, 2006, 8:00PCT)

On Jun 27, 2006  16:10 -0700, Andrew Morton wrote:
> Theodore Tso <[email protected]> wrote:
> > A similar split of the fs/jbd will be made in order to
> > support 64-bit jbd, which will be used by fs/ext4 and future versions
> > of ocfs2.
> 
> That's a bit sad.  I'd have thought that the JBD codebase could remain
> unforked - use a typedef and some kbuild trickery.
> 
> Or is something more that 64-bit-conversion being proposed?

One thing that a developer at CFS is working on is making the U. Wisconsin
"checksummed journal commit block" patch production ready.  While this
isn't a huge change to the code (a few conditionals whether checksumming
is on), it is more invasive than the 64-bit journal code.

This patch was originally developed as part of the IRON filesystems paper.
What it does is compute a checksum of each transaction's blocks and
store it in the commit block of the transaction.  During recovery the
transaction's checksum is again computed and if it matches the commit
block's checksum the transaction was fully committed, and is otherwise
incomplete or corrupted and should not be replayed.

The primary benefit is that we don't need to have a separate sync write
for each transaction's blocks before the commit block is written.
In the IRON fs paper this gave an improvement up to 20% on metadata
workloads.  It also makes the journal recovery more robust, because we
won't potentially scribble garbage blocks into the filesystem if some
blocks in the journal are corrupted (but the commit block is not).

Cheers, Andreas
--
Andreas Dilger
Principal Software Engineer
Cluster File Systems, Inc.


Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642