Optimizing Performance v6.2

Various publication server configuration options are available to optimize the performance of multi-master replication systems.

Almost all publication server performance related configuration options for single-master replication systems are equally applicable to multi-master replication systems (except when they are database produce specific, such as for Oracle). See Optimizing Performance for a discussion of these options.

The publication server configuration options are set in the publication server configuration file. See Section Publication and Subscription Server Configuration Options for a detailed explanation of how to set the configuration options in this file.

In addition, for configuration options specifically applicable to publication databases configured with the log-based method of synchronization replication, see Section Log-Based Method of Synchronization Options.

The following are some additional configuration options applicable to multi-master replication systems only.

uniquenessConflictDetection

The uniquenessConflictDetection option determines if uniqueness conflict needs to be detected at data load time or should be deferred to when data is applied against a target primary node. Possible values are EAGER and LAZY. Set it to EAGER if there is a high probability of duplicate inserts across primary nodes.

When the number of primary nodes is equal to two, then the conflict detection is performed in the default LAZY mode.

When the number of primary nodes is greater than two, then the conflict detection is always performed in EAGER mode. (A LAZY mode setting is ignored.) This is primarily required to avoid removing the already replicated conflicted changes from a target node, which otherwise is an expensive option.

uniquenessConflictDetection={EAGER | LAZY}

The default value is LAZY when the number of primary nodes is two.

skipConflictDetection

The skipConflictDetection option controls whether or not to skip conflict detection during synchronization replication. The default is false and should be changed only when the probability of data conflict across primary nodes is zero. For example if each primary node operates on an independent set of data then turning on this option iMMRoves the replication time.

skipConflictDetection={true | false}

The default value is false.

deadlockRetryCount

In a multi-master replication system, if a deadlock is detected on a target primary node, the deadlockRetryCount option controls the number of times the publication server attempts to retry application of the changes in the current replication cycle after waiting for the number of milliseconds specified by deadlockWaitTime. Set deadlockRetryCount to 0 to turn off this option in which case the failed changes are attempted in the next replication cycle.

deadlockRetryCount=n

The default value for n is 1.

deadlockWaitTime

The deadlockWaitTime option is used with the deadlockRetryCount option to set the wait time in milliseconds before the publication server attempts to retry application of the changes on the target primary node.

deadlockRetryCount=n`

The default value for n is 1000.