We have two ASE clusters and due to an office relocation we would like to split
these into component parts. The configuration is as follows:
2 x AlphaServer 8200 cluster
2 x AlphaServer 8400 cluster
1 x SW800
2 x dual redundant HSZ40 controllers
disks & tapes....
Note: The SW800 is split in half - one for the AS8200 cluster and the other for
the AS8400.
Software being run is as follows:
Digital UNIX v3.2G (lots of patches)
ASE v1.3
SYBASE
Note: This excludes propriety software, which only relies on disk services, rather
than NFS or DRD services.
A few comments that may help understand the configuration:
root, /usr (/var in /usr) and swap are on local RAID arrays (RAID 5
via SWXCRs).
SYBASE databases are all raw partitions, except system images.
all database areas and filesystems are mirrored under LSM.
all filesystem are AdvFS on top of LSM, except root and swap.
As part of the move, it has been decided by "the powers that be" that we should
take one AS8200 and one AS8400 in one move and the rest with another move
on a subsequant date. What I need to know is what is the safest method of taking
these two systems apart and running them for a period of time (up to a month) in
a split configuration and then putting them back together. To my mind, there are
three possible methods of doing this:
i. fail everything over to one node, shutdown the other node and
remove the node physically, ie pretend that that node has died.
ii. remove the node from ASE before removing it physically, so
ASE doesn't even look for the node.
iii. completely take out ASE from system startup and place the ASE
startup scripts (with suitable modification) in another area so that
services can be started.
I am thinking of using the first option. This is because I have done this for a cluster
in a similar circumstances, but only for a few days while hardware was being shipped
to fix a problem.
The second option has the benefit of being "cleaner" in that ASE will be told to ignore
the other node, but I am led to believe that removing ASE from system startup as it
cannot run in a cluster with only one node. Is this the case?
The third option I think is the only completely safe option, but requires a large amount of
work removing ASE from the system and then reintroducing it. It also would have the
added problem of making sure that all devices are in the correct state prior to getting
ASE to add them back into itself.
Has anybody any suggestions about what method would be the best way to do this, or
any other ideas about how to do this? Before someone makes the comment, we are
committed to doing this. All protestations about it "not really being a good idea" have
been ignored. We've got to do this, so I'm turning to you guys to point me in the safest
direction.
Many thanks for your help,
-jon.
--
Jon Morgan <jmorgan_at_dircon.co.uk>
DEC Systems Specialist
JAIS Europe Ltd
___________________
Received on Tue Feb 17 1998 - 14:10:25 NZDT