npars failover

[penelope_1]

Hi -
I was wondering if anyone could let me know about their experience with ServiceGuard cluster failover between npars on the same Integrity superdome. I have read this is a supported solution from hp, but how often could a hardware issue affect the whole superdome where the failover would not happen.
Do people generally failover to another npar in the same Integrity superdome or have a separate server for failover purposes? I have had experience with the Alpha GS80 where the server has gone down and it has affected both hard paritions, even though this wasn't supposed to happen.

Thanks in advance for any comments.

[Ian Vaughan · Accepted solution]

Howdy,
Npars are made up of one or more cell boards.
Different Npars are made of different cell boards. An npar is considered to be a "server" in it's own right as it runs a completely standalone copy of the OS from it's own hardware.

Cell boards are electrically separate from each other in the cell based systems so are supported as member nodes in a serviceguard cluster.

Superdomes have dual powerfeeds into each cabinet and the component npars are almost as resilient as having serveral standalone servers. The advantage of npars are the flexibilty you have especially if combined with vpars as well. Vpars within the same npar are not suitable for primary and alternate nodes for a serviceguard package but vpars in different npars are.

The Superdome only has 3 SPOF's - The System backplane, the HUCB board and the UGUY board. Don't worry too much about these as I've never heard of one failing yet.
In fact Superdomes have to be configured around 54 possible cominations of cells and I/O cardcages to give a balance between maximum protection against failure and maximum performance over the internal architecture.

Phew - hope some of that was helpful.
If you have any specifics - please ask.
Kind regards
Ian

[Torsten.]

Just to add:

The most critical situation is a power outage for the computer room or any worst case like fire or water.

For this reason you should have 2 redundant locations and cluster between them in different machines.

BTW, it's unlikely to see a SPOF failure in the superdome.

[melvyn burnard]

Serviceguard clusters built on Npars or Vpars in the same complex are supported, but with the obvious concerns about SPOF's

A Superdome has 3 offical single points of failure, and they do occasionally have a failure although this is very rare.

This document may be of use to your, especially page 8
http://docs.hp.com/en/6033/HPServiceguardClusterConfig.pdf

[penelope_1]

great document - thanks melvyn. I heard from a vendor yesterday that we wouldn't need the serviceguard software for failover with the superdome - is there something else built in that provides HA other than serviceguard if an npar\vpar is to fail?? I thought the serviceguard software is what would handle failover. I have this running in another environment on rp5400's and didn't know if there were other HA options with the superdome failover other than using serviceguards software.

[melvyn burnard]

Well whoever told you that is incorrect.
There is nothing that is available that will allow a failover of an application between Npars/Vpars other than an HA product such as Serviceguard.

[penelope_1]

thank you for clarifying melvyn

[penelope_1]

Thanks for the replies