Re: 10g RAC design options

Hi,

The decision to go with UltraSparc or x86 or Opteron processors is a decision which you will have to take on entirely as your responsibility!

Yes, there have been some benchmarks that x86 is faster than UltraSparc IV but then a lot is dependant on the application(s) you deploy and so on. This is more of a capacity planning exercise.

UltraSparc is a proven architecture and you have support for it. But cost is a big factor.

Now, Linux is also a good option since it is a lot cheaper. Support is the issue. 10g RAC is mature (though it has problems) but largely scalable and reliable. If you are using 10g RAC, then you don't need VCS or any clusterware for that matter.

I am not convinced that RAC scales on Linux as much as it does on other UNIX flavours. Adding nodes is not the solution for scaling (not always). A box must be able to scale vertically (adding CPUs, memory, etc.) but then scalability is another ball game all together. Yet, i don't think Linux boxes scale beyond 8-CPU/64 GB per node, maybe i am wrong. Technically, you may be able to stick in 1024 CPUs and 1TB of Memory into a Linux Server but i don't think it will work.

Patching is not an issue with 10g RAC, it can be done online. Also, it depends on what scheme you choose for installing ORACLE_HOME (local on each node or CFS install).

You may want to undertake a study if your application is suited for RAC, etc. If you have performance and other design issues on your single instance, it will be worse on RAC. This is not because RAC is bad but because applications need to 'RAC aware'.

Think about database consolidation techniques rather than having so many instances on a single server.

good luck!
ski
10g RAC Received on Wed Jul 06 2005 - 23:22:57 CDT