Oracle clusterware

[Stefan Farrelly]

Hi all,

Oracle are pushing their clusterware software here and Im wondering if anyone is already running it instead of Serviceguard ? If so any problems with it ?

Im having trouble finding anyone running it. We use MC/SG and a bit of VGS but Clusterware is being pushed bigtime by Oracle and theyre even bundling it free but we want to check it out before we try.

Cheers,

Stefan

[Stefan Farrelly]

oops, that should say "a bit of VCS"

[Duncan Edmonstone]

The big question I've never had answered satisfactorily by Oracle on their Clusterware is how it manages 'hung' nodes as compared to Serviceguard.

Serviceguard has some (minor) components that run in kernel space rather than user space - the most notable being the kernel safety timer.

Every time cmcld gets some CPU time it resets the safety timer (to 30 seconds I believe). If the timer ever counts down to 0 then the box is TOC'd. You can see this by killing cmcld on a SG cluster and within a maximum of 30 seconds the system will TOC.

THe great thing about this is that no matter what else is happening on the box, we can detect a hung node and eject it from the cluster before any data corruption can result.

Now clusterware can't run in the kernel - It's all user space and as such I don't trust its hung node detection as much - never had a decent explanantion from Oracle of how they manage this.

2nd point is disk management - as far as I can tell CSS does nothing to assist with managing disk resources in a cluster UNLESS you're using Oracle DB in which case they want you to use ASM. For any other app, it looks like you have to script it.

Of course I've not dug into it very deeply, these points put me off early...

HTH

Duncan