Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 26 Jul 2000 19:01:58 -0700
From:      Kirk McKusick <mckusick@mckusick.com>
To:        Peter Jeremy <peter.jeremy@alcatel.com.au>
Cc:        arch@FreeBSD.ORG
Subject:   Re: Snapshots in the Fast Filesystem 
Message-ID:  <200007270201.TAA16258@beastie.mckusick.com>
In-Reply-To: Your message of "Mon, 17 Jul 2000 11:15:39 %2B1000." <00Jul17.111555est.115264@border.alcanet.com.au> 

next in thread | previous in thread | raw e-mail | index | archive | help
	Date:  Mon, 17 Jul 2000 11:15:39 +1000
	From: Peter Jeremy <peter.jeremy@alcatel.com.au>
	Subject: Re: Snapshots in the Fast Filesystem
	In-reply-to: <200007060342.UAA23667@beastie.mckusick.com>; from
	 mckusick@mckusick.com on Wed, Jul 05, 2000 at 08:42:18PM -0700
	To: Kirk McKusick <mckusick@mckusick.com>
	Cc: arch@FreeBSD.ORG

	On 2000-Jul-05 20:42:18 -0700,
	Kirk McKusick <mckusick@mckusick.com> wrote:
	>I have completed an initial implementation of snapshots for the
	>fast filesystem (UFS/FFS).

	Thank you very much.  Having used the snapshot mechanisms in Compaq's
	AdvFS, I can say that this will be wonderful.

	Having read the README that you committed, I have a few questions:

	Disk space overheads: Does creating a snapshot require any additional
	disk space?  I realize that updates will require additional space for
	the `before' image, but what other metadata overheads are there?

There is about a 0.1% disk space overhead to initially create a
snapshot. Thus on a 10Gb filesystem, the initial snapshot will
consume about 10Mb.

	Running out of space: Is there any special behaviour when an active FS
	with one or more snaphsots runs out of disk space?  Does the (oldest?)
	snapshot become invalid to free up space, or is the behaviour
	indistinguishable from running out of space without snapshots?  How
	does this interact with the UFS minfree boundary?

I cheated on this one :-) I allow the snapshots to consume space
from the minfree reserve. Once the limit is hit, most processes
will die with filesystem full errors, so the rate of change on the
filesystem comes to a halt.

	Speed: Is it possible to reduce the time during which FS activity is
	suspended?  As you point out, one use for snapshots is doing point-in-
	time backups (I do this currently with Compaq Tru64 AdvFS
	filesystems).  From past experience, I've found that Oracle is very
	sensitive to FS activity blocks - after about 1 sec, Oracle will abort
	if it's FS activity is blocked.  It's possible that other applications
	have similar behaviour.

	Peter

The bulk of the suspend time is spent flushing out all the dirty
buffers and metadata to get a consistent on-disk copy. I do not
know any way to make that go faster, nor do I know a way to do it
without suspending filesystem operations. Applications that are
sensitive to 1-second I/O suspensions are likely to be problematic.

	Kirk McKusick


To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-arch" in the body of the message




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200007270201.TAA16258>