Comparison between shared queues and cluster queues

This information is designed to help you compare shared queues and cluster queues, and decide which might be more suitable for the system.

Channel Initiator costs

In cluster queues, messages are sent by channels, so allow for channel initiator costs in addition to application costs. There are costs in the network because channels get and put messages. These costs are not present with shared queues, which therefore use less processing power than cluster queues when moving messages between queue managers in a queue sharing group.

Availability of messages

When putting to a queue, cluster queues send the message to one of the queue managers with active channels connected to your queue manager. On the remote queue manager, if applications used to process the messages are not working, the messages are not processed and wait until the applications start. Similarly, if a queue manager is shut down, any messages on the queue manager are not made available until the queue manager restarts. These instances show lower message availability than when using shared queues.

When using shared queues, any application in the queue sharing group can get messages that are sent. If you shut down one queue manager in the queue sharing group, messages are available to the other queue managers, providing higher message availability than when using cluster queues.

Capacity

A coupling facility is more expensive than a disk; therefore the cost of storing 1,000,000 messages in a local queue is lower than having a coupling facility with enough capacity to store the same number of messages.

Send to other queue managers

Shared-queue messages are only available within a queue sharing group. To use a queue manager outside of the queue sharing group, we must use channels. We can use clustering to workload balance between multiple remote distributed queue managers.

Workload balancing

We can use clustering to give weight to which channels and queue managers get a proportion of the messages sent. For example, we can send 60% of messages to one queue manager, and 40% of messages to another queue manager. This instance does not depend on the ability of the remote queue manager to process work. The system with the first queue manager might be overloaded, and the system with the second queue manager might be idle, but most of the messages still go to the first queue manager.

With shared queues, two CICS systems can get messages. If one system is overloaded, the other system takes over most the workload.