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

RE: EXT3 and disk write back cache



Hi Stephen,

Thanks for the responses.  We have seen file system corruptions when we
conduct power cycling experiments.  Your feedback confirmed our findings.

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.  SR5 provides the following guarantee: suppose
blocks are written by an application in the order B_i, B_{i+1}, B_{i+2}, ...
to the block device exposed by SR5; upon recovery after an unclean shutdown,
if SR5 declares B_j as the last block written to disk media, then it is
guaranteed that block B_{j-1}, B_{j-2},... were already on the disk media.

We have performed extensive random power cycling experiments with EXT3 and
ReiserFS mounted on our SR5 device with disk write back cache on. The file
systems were not corrupted in these experiments.

You can learn more about SR5 at http://sr5tech.com.

Thanks,
Rob
------------------------------------------------------------
Boon Storage Technologies, Inc.
rgliu sr5tech com
(408)393-8998
http://www.sr5tech.com
------------------------------------------------------------
> -----Original Message-----
> From: Stephen C. Tweedie [mailto:sct redhat com]
> Sent: Thursday, February 20, 2003 3:35 PM
> To: rgliu sr5tech com
> Cc: ext3 users list
> Subject: Re: EXT3 and disk write back cache
>
>
> Hi,
>
> On Thu, 2003-02-20 at 22:39, Robert Liu wrote:
>
> > I have a few questions regarding EXT3 and disk write back cache:
> >
> > 1.	Will random power cycle cause data/metadata corruption on
> EXT3 running on
> > a hard disk with write back cache enabled?
>
> It might, unless the write-back caching still preserves the write
> ordering.
>
> > 2.	Will random power cycle cause EXT3 corruption when running
> on top of an
> > md RAID5 device with disk write back cache enabled?
>
> Software RAID5 will *increase* your sensitivity to writeback caching
> effects, not reduce it.
>
> > 3.	I understand Linux supports write barrier for IDE disks, does EXT3
> > utilize write barrier?  If so, will write barrier work for EXT3
> running on
> > top of an md device?
>
> There have been patches proposed for barriers, but none that work on an
> md device, and they aren't in 2.4 at present.
>
> Cheers,
>  Stephen
>





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