Multi-threaded optimization

You can enable multi-threaded optimization to improve the performance of Optimize, that is, shorten the amount of time it takes to run an Optimize session. To enable multi-threaded optimization, set the EnableMultithreading configuration property to true.

If you have multiple CPU or cores, the number of CPU or cores, therefore the maximum number of threads you can use may be restricted by your Optimize license. Contact your IBM® Unicarepresentative for more details.

Enabling multi-threaded optimization may significantly improve Optimize performance. The amount of performance improvement is dependent on many factors including the type of optimization rules you use, the interaction of those rules with your data, the speed of your database I/O, and your type and configuration of hardware. The maximum degree of parallelism achievable is governed by the non-parallel portions of processing and will differ from Optimize session to Optimize session.

In general, use the following guidelines.

Enabling multi-threaded optimization can only improve the performance of the optimization stage.

A significant portion of an optimization session run includes reading data from the Proposed Contacts Table (PCT), representing the optimization rules, preparing and retrieving data from the database, randomizing and creating chunks, and writing to the Optimized Contacts Table (OCT). Multi-threaded optimization does not affect these processes, which run serially. To improve the performance of reading the PCT and writing the OCT, consider using a database load utility and optimizing your database configuration.

If you do not use cross-customer rules in your optimization sessions, multi-threaded optimization has limited improvement.

Optimize leverages multi-threaded optimization most when processing the algorithms which implement cross-customer rules.

In general, the maximum number of threads available should be equal to or higher than the highest achievable degree of parallelism to maximize performance. However, you may find that other hardware limitations will reduce the performance leverage of multiple threads. For example, if insufficient RAM is available to support the number of threads in use, the optimization session may not be able to run. Or if your hardware has multiple cores (instead of CPUs) an multiple hardware threads use the same cache, your performance may be limited by the cache I/O.

You define the number of threads processing the cross-customer rules with the MaxCustomerSampleProcessingThreads configuration property. You may need to tune your system to find the optimal value of this setting based on your available RAM and hardware performance characteristics.

You may be able to run more threads in parallel if you have a smaller CustomerSampleSize because the RAM utilization per thread goes down. However, decreasing this value also decreases the amount of time required to process a chunk, decreasing the ration between chunk processing and preprocessing, causing the latter to become a bottleneck sooner.

If you have optimized the number of threads processing the cross-customer rules, you may be able to further increase performance by increasing the number of threads used reading data from a sample of customers or the threads used to write the optimized contacts to a staging table.

You define the number of threads reading data from a sample of customers with the ProcessingThreadQueueSize configuration property. You define the number of threads writing data to a staging table with the PostProcessingThreadQueueSize configuration property.

For more details about tuning multi-threaded optimization for your specific Optimize implementation, contact your IBM® Unicarepresentative.