Re: Anybody use Maximo on 8i OPS?

"ctemp" <ctemp01_at_hotmail.com> wrote in message news:4d44e6cf.0211141252.5c161098_at_posting.google.com...
> Hi,
>
> We are currently using Maximo 4.11 (Asset management s/w from MRO) on
> Windows 2000 and the backend is a single instance Oracle 8i database
> on Solaris 8. Since we are seeking a more stable and high avalaibility
> database, considering to upgrade the db to OPS which will be on Sun
> Cluster 3.0. But recently the Oracle guy told us that Maximo 411 won't
> work on OPS and he recommends us to use 9iRAC. The reason he said is
> that Maximo can not recognize the cluster environment and will cause
> problem. If we want to use OPS we have to modify the code in maximo.
> Also he said 9i RAC has no such problem. At this point I got confused.
> What maximo see the db on windows side is only a TNS connection and it
> should not matter if the db is a single instance or OPS and the
> failover thing is on the cluster. I thought all third part
> applications can work on either single instance or OPS db and there
> maybe such cases that some applications are not appropriate for using
> OPS in terms of performance. Maybe I am wrong. I had a kind of feeling
> that they are push us to use 9i RAC but maximo 4.11 is developed
> before 9i and not sure if it will be happy with 9i. Finally I asked
> him if we can use the HA option and he said 'yes'.

What your Oracle guy was getting at is that for Parallel Server to work properly, and not introduce a whole raft of new bottlenecks and points of contention, the application has to (or ought to) be 'partitioned'. I don't mean partitioning in the sense of range partitioning, or hash partitioning. I mean that it has to be re-written so that you try and arrange for particular chunks of data always to be accessed from one of the two (or more) instances running in OPS.

The reason for that is simply that in OPS, transferring data blocks between instances is a slow a nd painful experience. Say I, connected to Instance A, have queried the EMP table. Therefore, blocks of data from EMP are in my Instance's buffer cache. You connect to Instance B and query something from EMP. I therefore have to pass the EMP blocks from my Instance to yours -and I do so by writing them to disk, so that you can read them from disk. In effect, you've forced me to write blocks to disk. Forced writes like this is known as 'pinging' a block. And pinging blocks means it is very difficult to pass data around the cluster. So you write your applications with the aim of ensuring that you don't have to do it, or at least do it very infrequently.

You arrange, for example, that anybody querying the EMP table *must* establish a connection to Instance A, so the EMP blocks are always in that Instance alone. And if you want to query DEPT, you have to connect to Instance B, so the DEPT blocks never move from that Instance. And that's what we mean by talking about "application partitioning". It's a direct result of the massive cost of transferring data around an OPS cluster.

Now why do you think Oracle called the 9i version of OPS "Real Application Clusters"? Because in RAC, you can use "real" -that is, unpartitionedapplications.  What worked fine in a single instance database should work equally fine *without modification* on a 9i RAC. Why? Because in RAC, it is trivially cheap to move data around the cluster... we pass blocks of data via the high-speed interconnect between each machine, not via a slow hard disk. Block pinging goes away, so it now ceases to be a concern where anyone connects, or what data they want to access.

So that's what your Oracle guy was trying to say: that Maximo needs to be partitioned if its to work effectively on an 8i OPS, but that it can run unmodified on a 9i RAC. Without the modifications for OPS, the application would spend all its time writing to disk, and waits to get hold of data blocks would go through the roof. It's not that an application needs anything coded to make it "aware" it is running in a clustered environment... to that extent, you're quite right that to the app, it "should not matter if the db is a single instance or OPS". But a non-partitioned app in 8i OPS is going to run into trouble which it wouldn't were it running on 9i RAC.

Just to confuse the issue, it is actually a bit of marketing department hoo-har to say that RAC doesn't need applications to be partitioned. It's very largely true, but it's not entirely true: there are still costs involved in shunting data around over the cluster interconnect, and having to do that sort of work a lot can compromise application performance. There's an interesting paper by James Morle called "Unbreakable", available at www.oaktable.net (the 'Files' link) that discusses whether application partitioning really is unnecessary in a RAC environment.

But don't take that to mean RAC isn't a much better bet than OPS. It is. Hugely better. And your Oracle guy is perfectly correct in what he's telling you.

Regards
HJR
>
> Any comments will be highly appreciated especially for those who have
> experience on using Maximo.
>
> Thanks,
> Kevin
Received on Thu Nov 14 2002 - 15:25:18 CST