1. Replication Guide
  2. Improving Replication Performance

13 Improving Replication Performance

There are methods you can implement to improve replication performance.

The following describes methods for increasing replication performance:

  • Configure parallel replication. See Configuring Parallel Replication.

  • Use asynchronous replication, which is the default. See Making Decisions About Performance and Recovery Tradeoffs. However, if you are using active standby pairs, return twosafe (synchronous replication) has better performance than return receipt (semi-synchronous replication).

  • Set the LogFileSize and LogBufMB first connection attributes to their maximum values. See Setting Connection Attributes for Logging.

  • If the workload is heavy enough that replication sometimes falls behind, replicated changes must be captured from the transaction logs rather than from the in-memory log buffer. Using the fastest possible storage for the TimesTen transaction logs reduces I/O contention between transaction log flushing and replication capture and helps replication to catch up more quickly during periods of reduced workload. Consider using a high performance, cached disk array using a RAID-0 stripe across multiple fast disks or solid state storage.

  • Experiment with the number of connections to the database where the updates are applied. If you need more than 64 concurrent connections, set the Connections first connection attribute to a higher value. See Connections in Oracle TimesTen In-Memory Database Reference.

  • Adjust the transaction log buffer size and CPU power and resources. See Adjust Transaction Log Buffer Size and CPU.

  • There can be performance issues after altering tables with multiple partitions and extraneous space. See Performance Considerations When Altering Tables That Are Replicated.

  • Increase the number of threads that apply changes from the active master database to the standby master database by altering the RecoveryThreads first connection attribute. See Increase Replication Throughput for Active Standby Pairs.

  • Replication and XLA operations have significant overhead with transaction logging. Replication scales best when there are a limited number of transmitters or receivers. See Limit Replication Transmitters, Receivers, and XLA Readers.

Note:

Additional recommendations can be found in Poor Replication or XLA Performance in Oracle TimesTen In-Memory Database Monitoring and Troubleshooting Guide.