[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]
Re: Removing softdep
--- On Mon, 6/9/08, David Holland <dholland-tech%netbsd.org@localhost> wrote:
> From: David Holland <dholland-tech%netbsd.org@localhost>
> Subject: Re: Removing softdep
> To: "Gary Thorpe" <gathorpe79%yahoo.com@localhost>
> Cc: "Jared D. McNeill" <jmcneill%invisible.ca@localhost>, "Jukka Marin"
> <jmarin%embedtronics.fi@localhost>, "Martin Husemann"
> <martin%duskware.de@localhost>, "Andrew Doran" <ad%NetBSD.org@localhost>,
> Received: Monday, June 9, 2008, 12:03 PM
> On Sat, Jun 07, 2008 at 05:51:46PM -0700, Gary Thorpe wrote:
> > I don't think any software technique can help you
> if your drive
> > does write caching that the OS cannot control or at
> > influence.
> I'm not convinced of that, at least on a desktop,
> laptop, or
> non-heavily-loaded-server where the disk has a reasonable
> frequency of
> idle periods.
> David A. Holland
Both softdep and journaling assume that you can order at least some writes:
either using the dependency graph of softdep to ensure the meta data stays
coherent (that's my take on the theory behind it) or using a write-ahead log as
in journaling. If, in fact, the drive is not writing certain blocks to disk
when it says it is but is instead caching it, how do you think either of these
(or any other software scheme) will fare? How good will the journal be if it is
not actually updated before the operation it is supposed to be logging?
Someone else mentioned turning the write-cache off (which severely reduces disk
performance in some case) but for disks with tagged command queueing that works
properly, you can leave it on (most commodity disks do not have tagged command
queueing--that is for SCSI/SAS or higher-end SATA).
Connect with friends from any web browser - no download required. Try the new
Yahoo! Canada Messenger for the Web BETA at
Main Index |
Thread Index |