3-Node Oracle 9i RAC on Linux

[Sid_16]

Hi Folks,
We are in the process of installing a two-node Oracle 9i RAC on Red Hat
Linux AS on Itanium server. This also happens to be our first RAC implementation on Itanium Linux.

We would like to know, whether adding a third node to the cluster will cause any problem. We have seen success stories with 2/4 nodes in a
RAC. However, we have not seen the ones with odd numbers.

Anybody here who has done a 3-Node Oracle 9i RAC on Itanium Linux ? Were these any issues ?

Any info on the same would be appreciated.

TIA
Sid

[Jeroen Peereboom]

Sid,

Currently I have a 2-node cluster for test purposes. The official plan is for a 3-node RAC cluster and has been written by an Oracle consultant. He did not mention any problems with 3 nodes.

JP.

[Julio Yamawaki]

Hi,

There area no problems with odd number of servers in a 9i RAC configuration.
You MUST follow all pre-reqs and you won´t have any kind of problems.
Current, we have a pre-prod cluster with RH ES 3 Linux and Oracle 9i with 3 servers and MSA 1000 storage.
I think the problem of more then 2 servers is the connections to storage. If you use SCSI this can be a problem

[Hein van den Heuvel]

There is absolutely no problem adding a third node. Personally I like 3 nodes as is it the minimum needed to have low impact reduced availability.
If one of 2 nodes, each running at 60% usage average, fail then the remain needs to do 120% worth of work. The users will notice.
If one of 3 nodes, each running at 60%, fails then if you take one down the remaining 2 must deliver 90% each. Minor impact.
In the 4 node case you can load up to 70% average and still have a little head room to spare if one fails and the remaining boxes take must produce just over 90% each.

Some folks at Oracle (RAC designers) I worked with are convinced that RAC scales linearly only after 3 nodes, as for data contention resolution at the most 3 nodes will get involved, no matter how many nodes there are in the RAC cluster. I have a hard time believing/mearusing that.
In very course numbers, and application dependend of course, one could speculate the following numbers:
Assume 1 node with no rac delivers 100% work.
Switch on RAC and drop to 95%.
Add a second node and you add 'only' 90% more work power.
The third node and beyond would then add 85% of the base performance for eache node added.
The numbers are made up, but based on some RAC observations.

hth,
Hein.