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

On Tue, 22 Jul 2003 15:01:10 -0400, Ryan <rgaffuri_at_cox.net> wrote:

> why would implementing RAC increase I/O? is that because of the 2 phase
> commit? So you have to write to multiple datafiles?

No.

It may increase I/O for a number of reasons. A buffer in Instance A may have to be flushed for no other reason than Instance B has a past image of that buffer, and *it* needs to over-write its contents. The problem of forced disk writes (one instance making another instance do some I/O it wouldn't necessarily want to do on its own) has not been abolished in RAC (but it's better than it was in OPS).

Secondly, the interconnect is not of infinite bandwidth, therefore reserving it for the transport of buffers which really, really need to be shipped to other instances in the shortest possible time may be quite important. Therefore, you use GC_FILES_TO_LOCKS to switch off cache fusion transfer for specified datafiles... at which point, you are right back to the 'flush to disk from instance A and read into buffer cache at Instance B' forced disk write activity which characterised Parallel Server.

I'm sure there are other reasons, but those two are the biggest, I suspect.

~QM Received on Tue Jul 22 2003 - 15:29:59 CDT