Jim wrote:
>Hmm... here is a URL to linux kernel list discussion on why Linus says
>"dump is depreceated":
>
>http://hypermail.idiosynkrasia.net/linux-kernel/archived/2002/week28/0429.h=
>tml
>
>I never saw problems on 2.2 with lage FS though. Guess it was just a
>matter of time.
I've heard of similar kinds of problems with dump/restore and Unix
(SunOS) systems as well. The problem is that dump/restore typically
thinks it 'knows' something about the internals of filesystems (both
on disk and how the kernel caches things in memory as well). If the
filesystem being backed up is mounted read/write, you have the
potential for inherent race conditions between normal user
application/kernel activities and dumps attempt to make a consistent
backup of the filesystem. Even if a particular kernel/dump&restore
revision level works correctly, there is a danger that continuing
development of the kernel may cause things to break under some
conditions. You may (and I have) gotten away with not worrying about
this; but it can happen.
Fundamentally, I don't think that a truly consistent backup system for
live filesystems can be developed without explicit mechanisms for
dealing with the issue of race conditions. Personally, I love the
idea of filesystem 'snapshots'; (which I first heard about with
Network Appliance boxes something like a decade ago). It's not only a
nice feature to have for other reasons, but basically makes this issue
go away. I can't wait until OS all vendors incorporate this
functionality into their filesystems by default. Unfortunately, I
think that Network Applicances might have patents related to this.
I'm counting the days....
Bill Bogstad
[EMAIL PROTECTED]
---
Send mail for the `bblisa' mailing list to `[EMAIL PROTECTED]'.
Mail administrative requests to `[EMAIL PROTECTED]'.
