Re: Oracle 10g RAC Question

rhugga_at_yahoo.com (Keg) wrote in message news:<6c795a35.0410010532.dac2002_at_posting.google.com>...

They are right everything is in the manuals, but I'm going to help you anyway.

> We are wanting to move from a large Sparc III based Sun box to 2 or
> more smaller AMD64 based systems using Oracle 10g RAC. (We currently
> run 9.2.0.4 w/o RAC) I know Solaris x86 64-bit is unreleased and also
> Oracle 10.x on Solaris x86 as well, but they should be within the next
> 12 months.

I would never recommend an architecture based on the planned releases of any software. You never know if the release will be available according to the plans (usually not) and you definitely do not know the quality of the prospective product. It should work other way around, prepare your future system based on experiences with the current product range. Never use anything fresh on a business critical system.

> 1) Is 10g RAC basically the parrallel server of old? Thus I am
> assuming that all disk devices must be seen by all nodes in the
> cluster and also must be using raw disk devices for storage? (or even
> better, for example: Can node1 see disk group X, and node2 see disk
> group X & Y, and then node 3 see disk group Y & Z. Can 10g RAC
> function in that type of config? My purpose would be to segregate the
> different applications so that they use specific nodes in our RAC, and
> thus an SGA tuned for that application)

More or less RAC is the beast formerly know as a parallel server. You are assuming right about disk devices. However, all nodes must see the same datafiles, so node 1 & node 2 & node 3 must see at least X. The storage config is the same for all nodes on RAC. Instead of raw devices you can also use a certified clustered filesystem.

> 2) Does 10g give me the capability to bring nodes down 1 at a time for
> maintenance and keep the database online 24x7?

You can bring one node down at a time, but you can't do an Oracle related maintenance (patching) with any of your nodes up (all nodes must run with the same version of Oracle SW).

> 3) How does archive logging work, is there a 'master' node that will
> do the archiving or does each machine do archiving? (or can each
> machine do archiving for redundanacy purposes or will each machine
> generate it's own redo?)

Each machine is generating it's own redo and if you're using cross-archivation (you should use it, but it is not working very well at least in 9ir2) you have all redos from all nodes on all nodes. That means that on the first node you have redos from all others and vice versa.

> 4) Does Veritas's Storage Foundation for Oracle manage raw disk
> devices? (the current version of DBE 3.5 we are using cannot import
> the same disk group to more than 1 machine at a time, so I am assuming
> I can't use DBE for Oracle RAC, correct?)

You must use some cluster layer software (HACMP on AIXes for example, Legato cluster) to enable a concurrent access from several nodes at once. I do not have experiences with this particular product, but there is definitely some cluster software from Veritas, I just forgot its name.

> 5) Lastly, we are mainly wanting to move to RAC because our
> bottlenecks are in the SGA, and we have tuned to a point where tuning
> can't get anything more for us. We can throw more hardware at the
> system but won't be gaining much besides inflating
> hardware/software/support costs. Our SAN is so fast and our
> tablespaces are laid on onto so many dedicated volumes that we have
> moved the bottleneck into the SGA itself. We have a myriad of
> different applications and a myriad of different user usages so tuning
> the SGA was a long and tedious process, and we feel our only viable
> avenue for performance enhancement is Oracle RAC. (not to mention I
> can reduce TCO by a significant amount going to the AMD64 platform)
> So, am I correct in assumming that each node will have it's own SGA
> and thus can tune each one differently than the other?

My advice based on my experiences with RAC is never use RAC, if there is any way how to do the job without it! If you're unable to tune your single instance database, then on RAC you will not be able to tune it at all. It is true that you can use different SGAs parameters on each instance, but you will also encounter new challenges related to the cluster environment. So probably it is not going to help at all.

> 6) With Oracle RAC, is there a way to route certain transactions to
> certain nodes, (ie: Somehow cause application A to be handled by node
> 1, whereas application B might be handled by node 2, BUT, in the event
> of a node failure, 1 node can handle all applications) There would
> obviously be a huge performance hit in the event of a node failure but
> at least operations will continue.

Yes this is possible and works with no problems. You can select the primary address for each client, where it should connect to and the failover address where client's traffic go in the case of a failure of the primary.

--
Dusan Bolek
http://www.db-support.com

Email: spambin_at_seznam.cz
Pls add "Not Guilty" to the subject, otherwise your email will face an
unpleasant end as SPAM.