Questions on Oracle Server on IBM SP2

My company currently runs Oracle on Pyramid Nile SMP systems, but for internal political reasons we are likely to have to move to IBM hardware in the medium term. One of the options is the SP2 MPP machine.

I have a few questions on how Oracle works on these systems.

1 The SP2 uses a "shared nothing" architecture, in which each CPU node has its own disks which are not visible to other nodes. Oracle provides some kind of software layer on top of this to make it look like shared disk. Is this correct?

2. Again, because of the shared nothing architecture, Parallel Server is required to use Oracle on SP2. Is this correct?

3 Assuming Parallel Server is required, how is it used? I am familiar with how parallel server works in a clustered system, but am less clear about an MPP system.

3a Is their a named instance running on each node?

3b Do clients have to connect to a specific node and instance or is there some kind of resource controller which spreads connections among instances?

3c Similarly, is there a resource controller for batch work, or does work have to be allocated manually. On our current SMP machines we have 4 batch streams running against the same instance simultaneously and these are allocated CPU resources by the OS. It doesn't matter if we have 4 8 or 16 processors. The OS takes care of it. If we have to completely redesign the batch when we add new processors it would be incredibly difficult.

If none of these add-ons are available, and an SP2 just behaves like a giant cluster, then it seems the organisational overhead seems huge.

Answers, straight from the manuals, or from direct experience will be much appreciated.

Jim Smith                      jim_at_jimsmith.demon.co.uk
Oracle DBA