Date: Sat, 20 May 2000 10:31:06 -0700 (PDT) From: Matthew Dillon <dillon@apollo.backplane.com> To: Greg Lehey <grog@lemis.com> Cc: Matthew Reimer <mreimer@vpop.net>, current@FreeBSD.ORG Subject: Re: Possible Vinum RAID-5 problems? (was: panic: ffs_valloc: dup alloc) Message-ID: <200005201731.KAA45787@apollo.backplane.com> References: <lists.freebsd.current.20000519075536.A31215@cicely5.cicely.de> <lists.freebsd.current.20000519182044.A47558@freebie.lemis.com> <39256530.1FD33AA5@vpop.net> <200005191943.MAA40388@apollo.backplane.com> <20000520124936.E57670@freebie.lemis.com>
next in thread | previous in thread | raw e-mail | index | archive | help
:> Wait a sec... softupdates does not depend on write ordering.
:> Softupdates issues all non-conflicting writes in parallel and
:> doesn't care what order they are written to the disk. When
:> those complete, softupdates will then followup with all the
:> writes that depend on the original set.
:
:Hmm. Maybe I've misunderstood, then. It was my understanding that
:soft updates writes metadata with WRITE_ORDERED set, and thus avoids
:having to explicitly wait for completion. It's the WRITE_ORDERED that
:Vinum doesn't handle correctly. On a single disk, it's just a
:question of not sorting around a WRITE_ORDERED request. Vinum will
:keep the WRITE_ORDERED on a single disk, but it won't ensure that a
:request for the same volume, but which is destined for a different
:disk, will not be written until after all components of a prior
:WRITE_ORDERED request for that volume.
:
:Greg
Softupdates does not use bowrite(). In fact, the only place anyone
uses bowrite() is in the UFS directory code, and if softupdates is
turned on even those few bowrite()s are turned *OFF*.
Softupdates is very strict in how it issues I/O. There are two cases.
In the normal case softupdates does not issue dependant I/O
until the I/O associated with the dependancy itself completes. In the
second case, where the kernel forces a buffer to be flushed, softupdates
will issue I/O on a non-dependant version of the buffer and then
redirty the buffer with the dependant data.
-Matt
Matthew Dillon
<dillon@backplane.com>
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200005201731.KAA45787>