Can I shadow a system disk in LAN connected cluster?

[Neelmani Pandey]

Could anyone help me on this.

I have two nodes NODE A & NODE B located at different location.I am booting NODE A with $1$DKA0:[SYS0] and B with $2$DKA0:
Both the Server & disk are at different location.Only etherenet connectivity is there.
Can I enable shadowing between $1$DKA0: & $2$DKA0: when both the nodes are up.

OS version is openVMS V7.2-1

[Andy Bustamante]

>>>Can I enable shadowing between $1$DKA0: & $2$DKA0: when both the nodes are up.

It depends.

There are multiple ways to approach this depending on distance, licensing, network topology and storage configuration.

>>>Only etherenet connectivity is there

A bridged network which passes all network traffic or a TCP/IP only network? What are the latencies between sites? If you have a "modern" TCP/IP only connection, then you're out of luck for today.

There may be options to pass cluster traffic over TCP/IP in future, but that would require an operating system upgrade. Migrating to a fibre channel solution may allow disk replication, but that isn't OpenVMS volume shadowing and may or may not meet your requirements.

Andy Bustamante

[Volker Halle · Accepted solution]

No, not in your configuration.

When both nodes are 'up', this means that they have already booted from their local system disks ($1$DKA0: and $2$DKA0:). You can't enable shadowing on the system disk AFTER you've booted from that disk.

You could enable system disk shadowing on both disks individually, but those 2 disks can't be in the same system disk shadowset.

To boot from a system disk shadowset, the system first needs direct physical access to one of the valid member of the shadowset. Node B could only directly access $2$DKA0:, but this disk cannot be a valid member of a shadowset together with $1$DKA0 at that time, because node B is not yet running and MSCP-serving that member to node A.

Shadowed system disks must have their members locally accessible to the system, which is booting from this shadowset.

Volker.

[Jan van den Ende]

Neelmani,

>>>
Can I shadow a system disk in LAN connected cluster?
<<<

Yes, it can be done, and we did for many years (until we moved to SAN storage)

(later edited in: I doubt if this works with only ONE node per site, we always had at least 3 nodes)

First, the locations must be "close enough", such that (light-speed or electric-speed, depending) latency does not become a hindrance.

Secondly, you need to thoroughly understand the issues at play here. If the next exceeds your understanding, PLEASE postpone until you get the implications.

- get away from SYS0 for ALL cluster roots and remove SYS0 altogether. Whenever the absence of SYS0 leads to a problem, be VERY glad, because those are the situations where presence would lead to system disk corruption.

- set up one node to boot from, say, SYSA; and configure as an NI boot node.

- set up the system disk as a one-member shadow set.

- set up nodeB to boot AS A SATELLITE from SYSB.

- same for nodeC from SYSC ( etc, if needed)

- add a disk on the "other" to the shadow set
(and preferably an other one)

- Look at the boot information for nodeB and C; change the info for nodeA to be similar, and change nodeA's console boot info to be similar to nodeB and C's .

Be aware that any FULL CLUSTER REBOOT will require manual booting from the node's own root, and after the other nodes are up, a manual reboot from the network.

MAKE VERY SURE that there are VERY CLEAR and VERY PROMINENTLY PRESENT operating instructions about reboot, both single-node and full-cluster.

--- Discussing this way of operation with several (then) DEC OS and Storage guru's lead (waaaaaay back at the 1997 Paris Decus) to their not finding fault and giving the (unsupported) GO.

--- We started this at VMS V6.2; and rolling-upgraded up to V7.3-1; when we moved to SAN storage. We never did any full-cluster reboots once this was running. But we DID move from FDDI connect (backed by 10MB ethernet) to 100MB ethernet (backed by FDDI). We replaced (also rolling) all storage, storage controllers, and Alphas.

-----

To summarize: Yes, (3 nodes and up) it CAN be done, BUT: it is NOT easy to set up, nor maintain.
The rewards can be (and to us were) worth the effort in our view.

YMMV

Proost.

Have one on me.

jpe