Replication and RAC

I am looking at options for scaling up a replicated system and would appreciate any advice.

On each site there is currently one machine (CON1) which acts as the master. It holds system configuration data and is the point of update for this data.

The data is then repliicated to read-only snapshots on the other two machines (RO1 and RO2) - where they are accessed and read by the system processes. In this way RO1 and RO2 provide redundancy for the system config data for the processes.

If RO1 becomes unavailable a system process can redirect a configuration request at RO2. Obviously if CON1 dies, then the configuration cannot be changed but the system will carry on reading the current information from RO1 or 2 and continue without effect.

Now the client wishes to duplicate this architecture on a second, geographically separate site and have the second CON box be a redundant copy of CON1 - ie - having two points of update in a multi-master (peer-to-peer) environment.

<CRAP DIAGRAM TO FOLLOW>

         _______                                     _______
        | CON1  |                                    |  CON2 |
        | ______ |                                    |_______|
            |      |                                            |    |
           |        |                                          |      |
          |          |                                        |        |
 ____|__        | _______            _______        _______
|  RO1    |       |   RO2    |          | RO3     |       | RO4     |
|_______|       |_______|          |_______|       |_______|

My design options appear to be:

1. Change the whole six boxes to a single cluster. (The databases are 8i but can be upgraded to 9i). This does not seem possible given that the boxes are situated on geographically (international) separate sites as a RAC would need to read from a single file system.
2. Have two clusters at each site of three boxes - with peer-to-peer replication between the geographic sites. (I don't even know if this is possible but it seems like a good idea).
3. Have a six-way peer-to-peer replication. (Sounds like an admin nightmare and slow to boot).

Option 2 seems to be best if it is possible, as the peer-to-peer replication would be managable between two logical databases.

I would also be grateful if anybody could point me to any good sources for information regarding architecture design for distributed systems.

Thanks for your time.

Ray Received on Wed Sep 24 2003 - 02:37:13 CDT