Messaging is a crucial aspect of every Java application, especially for applications involving Enterprise Application integration (EAI) or separation of concerns, such us multi-tier WEB applications.
Messaging can be separated in two main categories, synchronous and asynchronous. With synchronous messaging the initiator of a conversation waits for a replay to each submitted request, on the other hand, in asynchronous messaging the initiator is not interested for the replay. The most common and efficient way for Java processes to communicate in a synchronous manner is through Remote Method Invocation (RMI). Asynchronous communication is mainly implemented using Java Messaging Service (JMS).
This post presents a design pattern concerning asynchronous communication between Java processes for low latency and high throughput applications.
As mentioned above JMS is considered to be the “de facto” standard for asynchronous application messaging. Nevertheless JMS introduces a noticeable increase in latency due to internal checks and procedures involved to the message exchange life-cycle (even for in memory brokers). Our preferred way to handle asynchronous messaging for achieving low latency and high throughput is :
- If persistence is mandatory, then the best approach is to use JMS persistent queues or topics
- If persistence is not mandatory then you should implement asynchronous messaging as follows :
- Message container should be a List, preferably the ArrayList implementation or a Map, preferably the HashMap Implementation
- Sender processes should perform synchronized access, using the synchronized block, to insert messages to the container
- Implement a pool of receiver processes that access the message container synchronously, using the synchronized block, and retract messages
- Messages can be implemented in many flavors, our preferred method is Plain Old Java Objects (POJOs) implementing the Externalizable interface so at to manually handle the serialization process.