Building a Scalable Enterprise Applications Using Asynchronous IO and SEDA Model

Building a Scalable Enterprise Applications Using Asynchronous IO and SEDA Model

This article presents one solution to overcome the scalability issues related to the enterprise applications that must support a mix of fast and long running business processes, or with great or small throughput.

Let's define a simple, sample scenario that would simulate this situation. We have a front end web application that receives requests through HTTP and routes them to different web service endpoints. One of the platforms that is queried from one of the web service end points responds very slow. As a consequence, we would have an overall low throughput over time. This is happening as the slow responding service keeps a worker thread from the web server thread pool busy, preventing other requests to be served accordingly to agreed SLAs.

There is a solution for this, which, though is not standardized yet, is implemented in a form or another in almost all the servlet containers: Jetty, Apache Jakarta Tomcat, and Weblogic. It is about asynchronous IO.

The key architecture components used in the solution proposed here are:

• asynchronous IO in servlet containers
• Staged event-driven architecture(SEDA) model

Asynchronous IO in Servlet Containers

Servlet containers became a good choice to implement high scalable applications upon the introduction of the java nio library -- which allowed the transition from thread per connection to thread per request.

Shortly this was proved not to be enough and the problems occurred when implementing Reverse Ajax-based applications. There is no mechanism currently present in servlet API that would allow for asynchronous delivery of data to the client. Currently Reverse Ajax is implemented in three modes:

• polling
• piggy back
• Comet

Read full article from Building a Scalable Enterprise Applications Using Asynchronous IO and SEDA Model