Re: multiple databases in a single RAC cluster
Date: Wed, 12 Nov 2008 08:41:53 -0500
The purpose of "grid" is subjective, IMO. *Your* purpose for making a database grid seems to be consolidation. For others, it is massive scalability. I would agree that if your reasons for considering this configuration are cost savings and server consolidation, then commodity-type servers are probably the best choice (certainly for cost savings).
I would also agree with the other advice you've gotten here regarding consolidation and multiple databases on a single cluster. The primary case where you'll encounter "issues" with multiple instances per node is the same as the single-node, multiple-instance case. That is, figuring out who/what is hogging all the CPU on the server. There's no resource management provided by Oracle outside of a single instance, so you have no way to stop someone in instance #1 from killing performance for instances #2-6 on the same server. I also agree that running multiple RAC databases in a single cluster is not significantly more complex than running one RAC database.
All that said, I don't think you'll find any whitepapers or documentation recommending or describing multiple databases in a RAC cluster. What I see most often are customers building RAC clusters to support a single, critical (critical from the standpoint of performance, availability or both) application. In that case, it is foolish to put multiple instances on each node as that allows the potential for something to influence the performance and/or availability of the "critical" application. However, for consolidation, I do see many customers putting multiple databases on their clusters.
I do recall the Dell MegaGrid project that touches on some of the things you're thinking about. I found their homepage at http://www.dell.com/megagrid. There's a deployment whitepaper there that might provide some interesting reading (I skimmed it and seems potentially applicable).
On Wed, Nov 12, 2008 at 8:02 AM, orcl <orcl_at_comcast.net> wrote:
> Typically when people think of rac/grid (when I think of it) Im thinking
> scaling horizontally with "inexpensive commodity type hardware"
> Going high end hardware with allot of databses in a rac configuration seems
> like it would defeat the purpose of "grid"
> ie it comes down to what one node can process and how much bandwidth the
> interconnect can handle.
> Thanks again and Im still looking for documents that support this type of
> rac cluster.