Re: Oracle 9.2 RAC on Win2K?

Bjørn Engsig wrote a paper on 9iRAC on Windows. It's on www.MiracleAS.dk under "Technical information" then a bit down. It was written after I forced poor Bjørn (who hates Microsoft, of course) to do a number of installs and troubleshooting on Windows :-).

Mogens

Mladen Gogala wrote:

On 2003.06.24 23:29, Branimir Petrovic wrote:

I need help with clearing up one conceptual issue:

My understanding is that with Oracle RAC one set of physical database files "sitting" in the "middle" - shared storage are being accessed by multiple Oracle instances running on multiple physical servers (nodes). All instances "attacking" (sharing) the very same set of data files at the same time. Right or wrong?

Right.

Provided the answer on the above question happens to be "yes" - I'd like to ask List Folks how feasible is to assemble and successfully
(smoothly?)

run 0.5-1 TB database, use Oracle RAC and high performance shared storage
(say SAN) served ("pumped") by a number of Windows 2K servers?

The "number" of Win2K servers I have on mind is at least 4 "beefy" (as beefy as it gets in Windows wrld) Win2K "boxes" each running Win2K AS with lots of RAM and at least 4 CPUs, with perspective of adding more later.

The company that I work for (Oxford Health Plans, http://www.oxhp.com) has several OPS configurations, one of which has 4 HP 9000/N nodes with hyperfabric (CLIC) interface. Unfortunately, we're running HP-UX, not Win2k.
Theoretically speaking, having a windoze cluster is possible, although the reputation of Win2k, especially when it comes to the stability, isn't the one that I would like for a fault tolerant and stable system. Fault tolerance and stability are the primary reasons for considering RAC.

It would be nice (for me) to know if new nodes can be added to the cluster at any later time to improve performance (in order to deal with increase in usage or to accomodate growth over period of time).

Performance of a clustered database is not bound by the CPU power alone, IO throughput is much more important and limiting factor then sheer CPU speed.
If you want performance, my advice is to go with a bigger box. Having in mind the price of SAN and RAC licenses, you'll be much better of with one 16 CPU machine then with a cluster of 4 machines with 4 CPUs each. To answer your question, yes, nodes can be added at later stage.

It would be very nice to know if number of nodes is or is not limited
(otherwise than by raw I/O capabilities of the shared storage).

I believe that the limit for windoze is 8 machines per cluster. With Unix, I believe that the limit is 128.

Has anyone seen/run/stumbled over similar beast, if so - does it "fly" or it "stinks"?

Depends on the application system. RAC is a bit more tolerant, but you still want to minimize the block exchange between the nodes. Application system must be "functionally partitioned", meaning that related data should be accessed from the same node. Example: let's imagine that we have health care application system which has elements like members/groups/providers, claims and drug utilization/ 3rd party vendors. Each of the listed groups is a complex application system accessing several hundreds of related tables. Good thing would be to put each of the listed application groups on its own node with the 4th node reserved for batch processing. If block exchange is frequent and DLM has a lot of work to do, then "flying" means "free falling until you hit the ground".

Thanks (for any help, hints, links, etc.),

Branimir

P.S.

I've looked at number of metalink articles and found none yet to "scratch" this specific "itch" of mine.
Oracle RAC on Win2K is for some bizarre reason REQUIREMENT.

Requirement by who? It usually helps to make a business case for a specific configuration and benchmark it to see whether the performance is satisfactory. Let me suggest a little benchmark: 4 clustered, beefed up PC boxes vs. a single, 16-CPU IBM P690 with the latest "960" CPUs and AIX 5.2. You can benchmark price and performance. I'd be very intrigued to find out the outcome. It is very hard to predict it properly (wink, wink).

--
Please see the official ORACLE-L FAQ: http://www.orafaq.net
--
Author: =?ISO-8859-1?Q?Mogens_N=F8rgaard?=
 INET: [EMAIL PROTECTED]
Fat City Network Services    -- 858-538-5051 http://www.fatcity.com
San Diego, California        -- Mailing list and web hosting services
---------------------------------------------------------------------
To REMOVE yourself from this mailing list, send an E-Mail message
to: [EMAIL PROTECTED] (note EXACT spelling of 'ListGuru') and in
the message BODY, include a line containing: UNSUB ORACLE-L


(or the name of mailing list you want to be removed from).  You may

also send the HELP command for other information (like subscribing).