Re: few doubts about RAC

Jaffar,

You don't need to be RAC guru. Just RTFM as suggested. So just few quick comments inline:

> Q1. We are using midleware software 'texudo' to connect our databases.
So do we. There is nothing we use in tuxedo for Oracle connections. In fact we are moving away from it. No idea if Tuxedo has something specific for Oracle, but middleware is the right way to go for load balancing of database most of the time.

> My question is, if one of my node get down, or fails, will be my
> activie session or connections automatically redirects to another
> node?

Yes if you configure it to use TAF - Transparent Application Failover. Don't let it confuse you - it's not always transparent. After few zears we still couldn=E4t get it working 100%.

> If yes, then, if I am doing some query and in middle of my query
> I get this problem, then the other node will carry out from where
> other node got disconnected or it starts from frist again? what
> happens if my session doing DML?

If you do it in the middle of a query - it will resume on another node. However it=E4s done on the client side - the query will be reexecuted (completely from scratch) and ALL results will be shipped to the client. The client (OCI layer) will filter out the rows that were already fetched and give you only new ones. Don't remeber what happens when query returns different results. Theoretically your client should not see failover, but remember that you will have overhead of rerunning of the query.
If you are in the middle of transaction - than it's not transparrent at all. You session will give you errors at any call until you rollback.
Note, that you will also lose any session state (like current_schema or package variables).
These answers based on 9i. AFAIK, there is no difference on 10.1.

> Q2. Can I have two nodes with different configuration in terms of
> hardware , like cpus and memory or both nodes should be idential?
Yes, but platform must be the same (like no HP-UX + Solaris).

> Q3. How RAC will maintain the load balancing? Will it distribute the
> load parallely or distribute the load according to the node capacity?
In 9i load balancing sucks. The listener selects which instance to connect you to based on CPU consumption of the node. So if it's not your bottleneck - you are in troubles... or if your workload is very volatile... or if you have many instance on one node (like we do)... or you have persistent connections (like we do)... etc. In 10g I heard there is a new feature that allows better load balancing, but I cannot comment on it - RTFM.

--=20
Best regards,
Alex Gorbachev

--
http://www.freelists.org/webpage/oracle-l