Re: Contention on sequence numbers in parallel server

From: "Patrick Nobels" >>
....... Initially, the application was not written for Oracle Parallel Server. As a result of this I get heavy pinging of databaseblocks to synchronize the instances.

One of the major problems, certainly not the only one, are the sequences used in the application. Transactions on one instance very often have to wait for a transaction on another instance to complete. <<

Each instance should have its own cursor cache so you should not have this problem unless you built the sequences with nocache. Have you checked your init.ora, sequence_cache_entries, for how many different sequences you have allowed tobe cached at one time? I suggest one for every cursor that has cache in sys.dba_sequences.

Are you using the gc_files_to_locks parameter? We found this to be a necessity. If you are not using it your locks are being assigned in a round robin fashion (but you probably already know this.)

Out of curriousity, what is your OPS ping ratio?

Mark Powell -- Oracle 7 Certified DBA
- The only advice that counts is the advice that you follow so follow your own advice - Received on Mon Jan 11 1999 - 19:35:34 CST