There are several ways to tune performance in mule. I've just finished a page on performance tuning in the Mule 2.x User Guide that walks through the available performance tuning options and provides formulas for calculating threads. Following is an excerpt of the high-level information from that page.
Overview
Essentially, a Mule application is a collaboration of a set of services. Messages are processed by services in three stages:
Tuning performance in Mule involves analyzing and improving these three stages for each service. You can start by applying the same tuning approach to all services and then further customize the tuning for each service as needed.
About Thread Pools
Each request that comes into Mule is processed on its own thread. A connector's receiver has a thread pool with a certain number of threads available to process requests on the inbound endpoints that use that connector.
If you are using synchronous processing, the same receiver thread will be used to carry the message all the way through Mule, whereas if you are doing asynchronous processing, the receiver thread is used only to carry the message to the component, at which point the message is transferred to a component thread, and the receiver thread is released back into the receiver thread pool so it can carry another message. After the component has finished processing an asynchronous message, it is transferred to a dispatcher thread and is sent on its way.
Therefore, the receiver, component, and dispatcher all have separate thread pools that are in use during asynchronous processing, whereas only the receiver thread pool is in use for synchronous processing.
About Threading Profiles
The threading profile specifies how the thread pools behave in Mule. You specify a separate threading profile for each receiver thread pool, component thread pool, and dispatcher thread pool. The most important setting of each is maxThreadsActive, which specifies how many threads are in the thread pool.
About Pooling Profiles
Unlike singleton components, pooled components each have a component pool, which contains multiple instances of the component to handle simultaneous incoming requests. A service's pooling profile configures its component pool. The most important setting is maxActive, which specifies the maximum number of instances of the component that Mule will create to handle simultaneous requests. Note that this number should be the same as the maxThreadsActive setting on the receiver thread pool, so that you have enough component instances available to handle the threads. You can use Mule HQ to monitor your component pools and see the maximum number of components you've used from the pool to help you tune the number of components and threads.
Calculating Threads
So how do you calculate the number of threads to set? There are several factors to consider, including concurrent user requests, processing time, response time, and timeout time. All of these factors are described in detail on the Performance Tuning page, along with formulas you can use to determine the number of threads to set for the receiver, service, component, and dispatcher, and the number of component instances to configure. For complete information, click here.
