The original question is at the end of this email.
Continuous access does not provide seamless failover for a TruCluster. All
worst case failures would require a reboot of the cluster.
Suggestions for DR/HA included:
- Use LSM for everything but CFS root, usr, var and boot partitions. Clone
the cluster for quick recovery of the cluster upon EVA failure.
- Script failover to remap the LUNS and reboot
- Look at possible McData (or other vendor) switch options that might
support some seamless failure. I haven't done this yet.
- Performance of CA also has been raised as an issue, at least in
synchronous mode. I don't know how to quantify this yet. Our physical
separation of the EVAs would be only about 2km.
Thanks to:
Jeffrey Hummel
Kelly Graham
Tom Webster
Raul Sossa
Danielle Georgette
Thomas Sjolshagen
--On Wednesday, May 18, 2005 5:06 PM -0400 Todd Acheson <acheson_at_ohio.edu>
wrote:
> Hi,
>
> We had our EVA5K crash for unknown causes about 2 weeks ago and sustained
> 24 hours of downtime. This raised the issue of how we could build more
> redundancy into our storage stack.
>
> There is Continuous Access (CA), and as I understand it there would not
> be seamless failover. You would mount the LUNs from the active EVA and
> if that unit failed, you would have to manually remap all filesets to the
> LUNs presented from the backup EVA.
>
> I was pointed to HP MetroClusters as a way to seamlessly have EVA
> failover, but in looking at the HP website, Metroclusters only appear to
> work with HP-UX. Can anyone validate this??
>
> If CA is indeed only a manual failover for TruClusters then perhaps I
> should just go back to LSM and use it to maintain consistency between 2
> EVAs. I used to use LSM prior to TruClusters but am a little concerned
> because we have on the order of 20 ADVFS domains with some that are very
> busy. Not sure if LSM would introduce any performance penalties.
>
> Todd Acheson
> Ohio University
>
Received on Tue May 24 2005 - 15:30:11 NZST