Re: OPS and MC Lock Manager

enem9276_at_my-dejanews.com >>
We are planning to implement OPS in 8.0.5 on HP9000 nodes using HP's MC-Lock Manager as the DLM.

What are the biggest problems we should look out for in implementing OPS <<<

I can not say anything about OPS on HP, I have worked with OPS on Pyramid and Sequent. Just make sure you allocate enough resources to the lock manager at the OS level. We have had very good performance from OPS.

I strongly recommend that you allocate locks at the file level with the gc_files_to_locks parameter (?sp) otherwise you get round robin locking.

It is advisable to partition your tables into tablespaces that mostly support access from one node . That is you should try to locate most of the users, especially updaters, to a set of tables like accounts payable so that they all access the rdbms from one node while another group of users who mostly access say materials are on another node and the application tables for AP are in one set of datafiles and the material tables are in another. This type of arrangement will reduce pinging and you can find this recommendation in the Parallel Server Concepts Manual for 7.3.

Common tables should be allocated to another tablespace, if you have enough of them. If you can not separate your users and tables the odds are you will still be OK. But you will need to study the gc parameters in the reference manual as you may have to tune them a little bit more carefully if you can not separate the tables and users.

Good Luck. Oh, yes the OPS Concepts manual also has several tuning queries in it.

.
Mark Powell -- Oracle 7 Certified DBA
- The only advice that counts is the advice that you follow so follow your own advice - Received on Thu Jan 07 1999 - 19:19:42 CST