Considering performance for timeout flows

When you design timeout flows, the decisions that you make can affect the performance of your integration nodes and applications.

You can use the timeout nodes TimeoutControl and TimeoutNotification in your message flows to control the way in which your message flows operate:

  • Set the Operation Mode property of the TimeoutNotification node to Automatic. This setting causes a flow to be invoked at the interval that you specify in the Timeout Value property. If the downstream processing is intensive, and the flow is still busy when the next timeout occurs, the TimeoutNotification node does not send a message immediately. When the flow completes current processing, all timeouts that were missed while the flow was busy are processed individually.

    The value of the Additional Instances property of the message flow is ignored downstream of a TimeoutNotification node, and you cannot use this property to change the behavior of the flow.

  • Use two associated flows to perform user-defined timeout processing. Set a timeout with a TimeoutControl node, and notify the flow using a TimeoutNotification node (which behaves like an input node to start a new message flow thread). If the downstream processing from the TimeoutNotification node is significant, requests that are set up in the TimeoutControl node can build up. These requests are subsequently processed individually by the TimeoutNotification node in a single thread.

    You cannot increase the Additional Instances property of the message flow if it starts with a TimeoutNotification node, therefore you cannot apply more threads to increase the capacity of the flow.

Although you can use a TimeoutNotification node to cause nodes in a message flow to poll for the next item of work, this approach forces a delay between each transaction, and typically does not provide an efficient solution. If you want to periodically check a resource for the next piece of work, and process it immediately, consider one or more of the following alternative solutions:

  • Use a built-in input node.
  • Write your own input node by using the user-defined node API (in Java™ or C).
  • Consider purchasing an IBM® or vendor-provided adapter which polls the subsystem you want, and triggers the flow.

A message flow that uses these options can process more work overall than it can if you implement a timeout solution, and incurs lower CPU cost, although your initial development costs might be slightly higher.