|
>From jsc3 Fri Sep 1 17:53:51 2000
Subject: Re: NFS Redudancy
To: jeremy@xxxxxxxxxxxx (Jeremy Hansen)
Date: Fri, 1 Sep 2000 17:53:51 -0400 (EDT)
In-Reply-To: <Pine.LNX.4.21.0009011604020.28600-100000@xxxxxxxxxxxxxxxxxx> from
"Jeremy Hansen" at Sep 01, 2000 04:06:53 PM
X-Mailer: ELM [version 2.5 PL1]
Content-Length: 1447
It works as long as the filesystems are identical. That means either
readonly content dd'd to identical disk on failover machine, or dual
ported disk storage with two hosts (or more) attached. When the failover
happens the backup system then mounts the disks and takes over. You
need to be VERY sure that the primary system is NOT up and writing to
the disk or you will have to go to backups after the disk gets corrupted
(having two systems perform unsynchronized writes to the same filesystem
is not a good idea).
This is the way Sun handles it in their HA cluster. They go a step
further by using Veritas Volume Manager, and Veritas has to forcibly
import the disk group to the backup when a failover is done - the
backup also sends a break or poweroff command to the primary via the
serial terminal interface to make darn sure the primary is down.
That said, I have seen three systems all mounting the same volume
during some pathological testing I did on some systems at a Sun
HA training course. The storaage used in this situation was
Sun D1000/A1000 (dual ported UltraSCSI) and Sun A5000 (Fiberchannel).
> I've never tried this myself, but I'm curious, can nfs handle picking up
> client machine in a failover situation. If my primary nfs server dies,
> and my secondary takes over the first, can nfs clients handle this?
> Something tells me nfs wouldn't be very happy in this situation.
>
> Thanks
> -jeremy
--
John Cronin
--
John Cronin
|
