Is 10g RAC ready for OLTP

Question

To achieve high availability by eliminating single point of failures in the following areas we are thinking of having our OLTP DBs on a single 10g RAC cluster
i) OS/Firmware patches requiring reboots
ii) Unplanned server failures
iii) One-off Oracle patches
We have migrated our DSS systems to 10g RAC (windows x64). However, in the last 9 months since we deployed we have seen 2 issues: single node eviction & multiple node evictions. Single node eviction is supposedly fixed w/ a patch that needs clusterwide shutdown.
The baseline for me on OLTP is 8i where I have taken DT once in 2 years to apply Oracle patches, once a year for OS patches and very rare server failures resulting in DB failover.
Questions I have:
a) Is 10g RAC really stable to be used for OLTP?
b) How is this being designed elsewhere with a view to reduce planned/unplanned DTs
thanks,
SM

Answer 1

> a) Is 10g RAC really stable to be used for OLTP?
Loaded question as you are implying that until now, RAC has not been stable and not robust enough for OLTP.
Stability for any system is dependent on:
- platform h/w
- storage h/w
- network h/w
- o/s s/w
- application s/w
- administration
What RAC buys you is having multiple database instance for a single physical database. Which means that in the worse case where you are forced to down a platform because of one of the above reasons, the remaining platforms in the cluster should still be available.. courtesy of the share everything approach.
But RAC alone is not the answer.. there are numerous factors to consider. One of my longest uptime databases is an Oracle SE server with a 12,000+ uptime. And it is used 24x7 as a data collection platform.
It went down recently. The cause? Network errors and power failures that resulted in the rack cabinet housing this server, to be reset.
Have numerous examples of how unforseen events caused disaster in a computer room. From dirty electrical power to an aircon automated switchover failing.
RAC does not solve any of these. What happens when there is a power failure or h/w error with the switch used for the Interconnect? Without the nodes being able to communicate with one another, all nodes will be evict themselves from the cluster.
Looking at RAC alone as The Solution to your H/A requirement is a bit naive IMO. Yes, RAC is an excellent and major cog in the wheel of H/A.. but there are others too.
Q. Is 10g RAC really stable to be used for OLTP?
A. As stable and as robust as you make it to be.