Performance Tuning for Content Manager

176 Performance Tuning for Content Manager

Remember, lay out the plan first. Think through before you start changing the

content of the DDL. For your physical system layout, we recommend:

 Separate the operational system paging area from the other components.

 Separate the physical location of databases from each other.

 Separate the staging area from LBOSDATA.

 Separate the paging area from LBOSDATA.

 Separate the table space containers onto different disks.

For more information about DDLs, databases, table spaces, and containers, refer

to the publications listed at the beginning of this chapter.

8.2.3 Create separate instances during installation

The default Content Manager installation uses one database instance for all

databases. Therefore, if you install the Library Server and the Resource Manager

database on the same AIX system, you will have two databases managed by one

database instance (db2inst1):

 Library Server database (icmnlsdb)

 Resource Manager database (rmdb)

The advantage of this approach is that you avoid the administration work of

managing multiple instances. The potential disadvantages of the single-instance

approach are:

 Shared memory: A single DB2 instance on a 32-bit AIX can use only up to 1.7

GB of memory for database buffer pools, which means that all of the

databases would have to share this memory.

 Resource contention.

 Stoppages to the instance affect all resident databases instead of just the

intended database.

Because the database manager (instance) is a complete environment, the

parameter settings apply to every database in that instance. We recommend

creating one instance per database. For details about how to do this, refer to 6.7,

“Creating and configuring DB2 instances” on page 177 in Content Manager

Version 8.1 Migration Guide for Multiplatforms, SG24-6877.

8.2.4 Initial parameter tuning for Library Server database

You can improve performance when the database manager is started (cold

starts) by having a pool of worker agents created during startup. These initial idle

agents will be available to handle initial queries instead of being created at query

time. This improves the performance of those initial queries.