•By default, QSERVER subsystem (QUSRWRK on V5), runs in the base memory pool.
•Create a separate pool and adjust the subsystem description to run in this new pool.
•Adjust the prestart job entries so that they run in the new pool.
•This segregates the memory used by ODBC, JDBC and OLEDB into a separate space and reduces contention and fragmentation.
•Make sure you set the pool to *CALC with the WRKSRHPOOLS command so that it uses expert cache.