11.8 Miscellaneous Best Practices

I l @ ve RuBoard

Finally, here is a set of miscellaneous best practices that have not been mentioned anywhere else in the book. Keep these in mind when creating the various parts of your enterprise system.

11.8.1 Design

• Design objects so that they can be easily replaced by a faster implementation.

• Use interfaces and interface-like patterns (e.g., the Factory pattern).

• Design for reusable objects.

• Use stateless objects.

• Consider whether to optimize objects for update or for access.

• Minimize data conversions.

11.8.2 Monitoring

• Constantly monitor the running application.

• Retain performance logs. Choose one set as your comparison standard.

• Monitor as many parameters as possible throughout the system.

• Note every single change to the system.

• Listen to the application users, but double-check any reported problems.

• Ensure that caching effects do not skew the measurements of a reported problem.

• Train users to use the application efficiently .

11.8.3 Parallelism

• Design parallelism into the application wherever possible. Identify what cannot be parallelized. Watch out for too much parallelism. There are diminishing returns from parallelism overheads.

• Balance workloads. Unbalanced parallel activities can limit the performance of the system.

• Split up the data among many different files (preferably on separate disks).

• Support asynchronous communications.

• Decouple activities so that no activity is unnecessarily blocked by another activity.

• Minimize points where parallel activities are forced to converge.

11.8.4 Distributed Computing

• Minimize the communication between distributed components .

• Avoid generating distributed garbage.

• Reduce transfer costs by duplicating data.

• Cache distributed data wherever possible.

• Minimize the synchronization requirements of duplicated data.

• Use compression to reduce transfer time.

11.8.5 JMS

• Close resources (e.g., connections, session objects, producers , and consumers) when you're finished with them.

• Start the consumer before the producer so that the initial messages do not need to queue as they wait for the consumer.

• Use separate transactional sessions and nontransactional sessions.

• Use nonpersistent messages, as they are faster than persistent messages.

• Use shorter messages, as longer messages take longer to deliver and process.

• Specify the lowest redelivery count and time-to-live that the application can accept.

• Maximize message throughput by tuning the delivery capacity.

• Use asynchronous reception of messages with the MessageListener interface.

• Process messages simultaneously with ConnectionConsumers and ServerSessionPools .

• Throttle very high-volume message delivery to a rate the consumers can handle. Use load-balanced message queues if necessary.

• Duplicate-delivery mode is the fastest possible delivery mode, followed by auto-acknowledgment mode. Try to avoid client acknowledgment mode.

• Use publish-and-subscribe for large numbers of active listeners, and point-to-point connections for few active listeners.

I l @ ve RuBoard