[WBEL-users] Custering in WBEL/RHES/AS

[Pete Stevens]

On Mon, 25 Apr 2005, kirby wrote:
> Note:  30 min unplanned downtime is GREAT.  My experience with small
> clustered systems is that they cause more problems than they help.  It is
> VERY complicated to set-up.  Maintenance becomes a pain.  No one knows how
> to support it.  The technology is still bleeding edge.
> For a large cluster with sophisticated Unix admins and DBA's it can be good,
> but IMO a clustered db is about 3x the work for and extra (maybe) .001%
> uptime improvement.  If you are AMEX or VISA it is worth it... For most
> others.  I go for KISS.

For what it's worth I agree. My experience has been that more downtime is
caused by screwing up the configuration of the cluster, than occurs due to
hardware or other type of failure.

I achieve reliability by having a hot spare with an up to date (<1 minute
behind) backup. It's a manual process to bring the spare in but it allows
recovery from unplanned downtime in minutes, rather than hours or days. It
also allows you to break the mirroring during an upgrade so that you can
upgrade the spare, make it live, test it with instant fall back to the last
system before making them hot spares again.

Obviously this won't do if you've signed a 99.999% uptime contract, but 99.9%
is readily achievable with this method.

Yours,

Pete Stevens

--
Pete Stevens
pete at ex-parrot.com
http://www.ex-parrot.com/~pete/

        We are developing the social individualist meta-context for the future.
                                             -- Samizdata.net mission statement