Oracle and four node cluster

[Vladimir Fabecic]

Hello
I do not have much experience with clustering on linux.
Requirement:
3 servers have to run one or two Oracle instances (not RAC). If one node fails, 4th server has to take over the instance(s) of failed one.
Any suggestion how to make it would be welcome.
Is it possible to make active/active configuration, i.e. that every server run one instance and take over instance of failed server? If so, how?
Wcich cluster software to use?
Thanks

[Ivan Ferreira]

I know about Red Hat Cluster Services.

Using this software, you can configure the Oracle as required for your installation. Every node can run an instance of oracle and if the node fails, another node can start both instances.

You need a shared disk storage for this (For example MSA, EVA, etc). It's also recommended that you use a fence device for maintaining data integrity. Check the Red Hat Cluster Suite documentation for detailed information. The fence device can be for example, HP RILO available on most servers or a Brocade SAN Switch used to connect the storages.

Every node has access to the shared disk storage. If a node fails, as the other nodes have access to the storage also, you can configure to start the failed oracle instance in any other node.

[Steven E. Protter]

Shalom,

Active-Active is not possible without RAC. It would immediately corrupt the Oracle database on shared storage.

If the data was read only and hosted on local disk of each node, it would be possible to run oracle on all nodes. There would however be no failover under this scenario.

Red Hat Clustering or Service Guard for Linux can handle an active-backup clustering scenario for you.

SEP

[Ivan Ferreira]

What I mean by active-active is to run one instance in one node accessing to one disk partition, and another instance on another node using another disk partition (Not the same instance on both nodes). Each node is failover node for the other.

[Serviceguard for Linux]

Serviceguard for Linux can handle that configuration. A toolkit for Oracle failover is also available. The toolkit is licensed per cluster.

http://www.hp.com/go/sglx

[Vladimir Fabecic]

Thank you guys.
Looks like there is a good way to do it.
I have to replace TRU64 cluster with linux solution. With active/active I meant no dedicated standby server and cluster file system.
Can you post some links with documentation?
Which clustering software is better and which is cheaper?
Is there some "best practice" example?
Shared storage is EVA3000. Which Redhat distribution (or SUSE?) is the best choice?
Do I need secure path?

[Serviceguard for Linux]

The link I posted for Serviceguard for Linux is the best starting point. www.hp.com/info/sglx has much of the documentation. While it is a little dated, the configuration guide gives a lot of info on best practice for cluster configuration.

As for pricing, the list price is roughly $1300 per server, but check with your distributor or sales person for exact pricing. Various support options are available. The Oracle toolkit list is $995.

The latest full documentation is available as well:

http://docs.hp.com/en/ha.html#Serviceguard%20for%20Linux
or more specifiically:
http://docs.hp.com/en/B9903-90046/B9903-90046.pdf and
http://docs.hp.com/en/B9903-90048/B9903-90048.pdf

Hope this is a useful start

[Vitaly Karasik_1]

RH Cluster Suite costs about $500 annually.
I'll prefer RH Cluster Suite instead of SG because in this case you have one vendor and one support point == Red Hat. In addition SG uses (AFAIK) closed source kernel modules, so you cannot use the latest recommended kernel version, but must to wait when HP will release appropriate SG modules version.

As for securepath - you can use Qlogic driver failover.

[Vladimir Fabecic]

Hello Vitaly
Thanks for interesting data.
Vitaly and "Serviceguard for Linux", what are feature differences between RH cluster and HP Serviceguard cluster?
Also would like to use RedHat GFS. I have seen that it works OK with Serviceguard, so no problem with that.
Regards

[Serviceguard for Linux]

Correction to Vitaly comments.

Serviceguard does not use closed sourse kernel components. We have had no issues with support from RedHat or Novell(Suse).

Customers can buy RedHat or (I believe) SUSE from HP and get support from HP.

Serviceguard for Linux is based on Serviceguard for HP-UX. That means it has a history greater than 10 years and we are constantly looking for improvements. One thing we focus on is no single point of failure in the cluster. For any cluster you should look at every component within that cluster and see if its failure can possibly take down the cluster or worse.