Using serviceguard to provide nic failover.

[Ronnie Doggart]

We are installing Oracle 10GR2 and want to use provide resilience at the network interface level. The Oracle cluster software requires that a network interface is defined for its cluster interconnect. This is a single point of failure. Oracle however requires that the nic be identified via adapter:subnet e.g lan1:10.1.1.0.

How can we provide resilience on the nic via serviceguard ?

[IT_2007]

you can define standby IP address in cmclconf.ascii file and recompile the cluster. In case of primary network failure, it will automatically switch to standby NIC.

[Jeff Schussele]

Hi Ronnie,

You may also want to look at HP's APA (Auto Port Aggregration) Product.
It will allow you to "gang-up" multiple NICs and as long as they *all* don't fail you'll still have network access.
Also gives the added benefit of faster network speed - i.e. if you aggregrate 4 GigE cards your effective network speed would be 4 GB.

Rgds,
Jeff

[Ronnie Doggart]

We looked at APA's Lan Monitor capability which allows you to bond two physical nic's into a logical one, but it does not work under Serviceguard.

When serviceguard fails over from lan0 to lan1 does it redirect traffic from lan0 to lan1. Oracle's cluster software requires the adapter to be configured into the software. So how would serviceguard handle this?

Ronnie

[Mel Burslan]

You need to make a decision and use either Oracle's cluster software OR serviceguard. If you use both, I am not sure about the outcome when something happens to your network and both cluster software try to remedy the situation.

serviceguard, in case of a lan failure, experiences a short duration of hickup and switches to the secondary interface and keeps going under the normal circumstances.

[Shardha]

I agree with the option to define a unused NIC as a standby NIC and apply the configuration. This will give NIC failover. The package will not fail over to other node only the alias IP address will be moved to the standb NIC in case of first NIC fails.

[Ronnie Doggart]

I do not believe that Oracle's cluster software handles network failover. I have raised this question with Oracle and am currently waiting on a response.

We need to use serviceguard as we want to use shared logical volumes rather than raw devices and you need Serviceguard extensions for RAC to do this.

Is there anyone using Oracle 10GR2 and serviceguard in a RAC environment that has redundancy for the cluster interconnect nic.

Ronnie

[Mel Burslan]

I do not want to badmouth HP but for Oracle10 RAC implementations, you might want to seriously consider Veritas cluster in conjunction with their cluster filesystems. I do not have it in my datacenter but I am hearing they do a heck of a job for what you want.

I am not sure about if oracle clustering is able to handle network failover or not, but if it does not and you still want to use oracle clustering, your only option is to use HP APA for network redundancy.

FYI.

[IT_2007]

I totally agree with Mel. see his above post. We do have Oracle RAC in MC/Servicegaurd environemnt and we do use port aggregation. We do use logical and raw volumes and available read/write on both nodes same time (server and client).

[Ronnie Doggart]

IT_2007,

Do you use port aggregation for the cluster interconnect used by Oracle?

Could you show me the output from :

oifcfg getif

Ronnie