Balance Dispatcher Load with In-Progress Request Factor

The weighted round robin format of load balancing treats all requests as equal, and all dispatchers as equally capable of handling the number of requests that they receive. However, different requests require more or less processing power. Dispatchers also run on different servers, with different processing capabilities. For example, if a dispatcher falls behind because it is running on a slower server or because it is getting a lot of requests that require a lot of processing power, the round robin format still treats all dispatchers the same. Dispatchers that start to fall behind have a higher number of in-progress requests in their queue. The round robin algorithm can use this information to avoid sending new requests to those dispatchers until they're no longer overloaded.

The inProgressRequestFactor advanced setting controls how much feedback is sent to the round robin algorithm. The larger the value, the less likely it is that a node with more in-progress requests will be used. Our research shows that the ideal amount of feedback is the default value of 2.0. To use a simple round robin format, set it to 0.0 at a system level.

You can set the value at the system level or at the service level. The system level setting is used as the default for all services. The service settings take precedence over the system level setting.

You can also control dispatcher load balancing by setting capacity processing. See Balancing requests among dispatchers. You can also turn off the weighted round robin format of load balancing for the dispatcher. See Setting the dispatcher load balancing property to cluster compatible mode. You must have the required permissions to access IBM Cognos Administration functionality. See Secured Functions and Features.