UIMA AS currently provides no way to change prefetch size on a reply queue. A default of 1 is used. This slows down the rate of processing messages by a service. Each UIMA AS service has a dedicated submitter thread whose main job is to deliver a message to a thread pool for processing. Since the submitter thread executes very fast and the prefetch=1 , the submitter thread is constantly waiting for the Broker to deliver the next message. Increasing prefetch forces the Broker to deliver multiple messages to a service which buffers them until the submitter thread is ready. The size of the prefetch should be equal to a scalout factor defined on each reply queue. The default is still 1.
NOTE: increasing prefetch effects memory requirement for the service. The larger the prefetch the more messages are buffered taking up more memory.