Re: Upgrade to big SUN box or RAC for data warehouse?

Brian Peasland <dba_at_remove_spam.peasland.com> wrote in message news:<3F1D9CAD.F5A44E1D_at_remove_spam.peasland.com>...
> > why would implementing RAC increase I/O? is that because of the 2 phase
> > commit? So you have to write to multiple datafiles?

   Ideally, RAC would allow you to DECREASE disk IO. When you've got multiple SGA's on different nodes in the cluster, the goal would be to do block transfers between the various nods/SGAs rather than the disk subsystem. I've seen a disk bound OLTP system get a greater than 100% boost from RAC just by going to a 2nd node.

    However, you still should only try RAC last after you've done everything else you possibly can to tune your single instance.

>
> RAC by itself won't necessarily increase I/O. But the argument is
> similar to adding more CPUs to a single server instance. With more CPUs,
> processes get done with their computations quicker and therefore can
> turn to waiting for I/O more often. More processes waiting for I/O can
> cause bottlenecks in I/O. With RAC, you are increasing the number of
> processors, just not all on the same server.
>
> This doesn't have anything to do with 2PC (which is for distributed
> databases).
Received on Wed Jul 23 2003 - 13:23:43 CDT