|Bookshelf Home | Contents | Index | PDF|
Using AOM thread pooling can improve performance and scalability on multiple-CPU machines that are under heavy load—for example, machines using eight or more CPUs that are running at more than 75% CPU capacity.
AOM thread pooling reduces some of system resource usage devoted to creating and closing session threads, as users log in and log out or are timed out. As when you are not using thread pooling, session threads are created as needed as session requests demand. However, instead of being closed when a session terminates, they are released to a pool, where they become available for use by a subsequent session.
NOTE: Using thread pooling introduces its own overhead, however, such as in task context-switching. For this reason, it is strongly recommended not to try to pool threads without also multiplexing them (that is, do not set UseThreadPool = TRUE, but ThreadAffinity = TRUE).
Because ThreadAffinity = FALSE, threads are multiplexed, as can be done with certain types of database connections or SISNAPI connections. At any given time, each thread may be dedicated to one or more user session (task).
To determine an appropriate value for MinPoolThreads and MaxPoolThreads, start slowly, monitor system performance, then introduce more multiplexing, as may be appropriate for your deployment. For example, start with a formula like this (based on two tasks per thread):
|Performance Tuning Guide|