Sunday, March 19, 2006


Most web based J2EE application use an MVC framework like Struts, Velocity, Spring, etc. All these frameworks provide a Controller which intercepts requests from the client and determines the model element to invoke. Will this change once AJAX becomes more widely adopted? When we use AJAX, all requests first go to the AJAX engine which in turn makes an HTTP request to the server. The response is monitored by a callback function which acts upon the response when it is recieved.

A reason why an alternative architecture may emerge, is because of the shortcomings of using an MVC framework. Despite of the advantages that are promised by some of these frameworks, they do have many strings attached. Once you use a framework, you have to do things *their* way. They very often influence the application's architecture also. Making the architecture flexible enough to replace the framework has to be done by creating a layer of abstraction between the application and the framework. This adds to the complexity and cost of the application. Most MVC frameworks cannot be used without a rather steep learning curve. This problem is even more pronounced when a team has to invest time in learning different frameworks for different clients. I have known people to take as much as a month to get comfortable with Struts, which is quite a huge time investment.

However frameworks do have their advantages also, which is why they are probably so popular. The biggest advantage is reusing the plumbing code of the framework, and declarative programming which most frameworks support. Declarative programming however has it's own drawbacks. As the application grows in size the configuration files tend to become very unwieldy.

Frameworks are not necessarily bad. I think a balance has to be created between using a framework and creating our own code. Perhaps light weight frameworks will work best. I used a big time user of frameworks, but over the years I have realized that the productivity benefit of the plumbing code comes at the cost of a steep learning curve. This is particularly painful when we have to undergo the learning curve of an entire framework for using just a part of it.

An architecture that might emerge is one in which the Controller resides on the client (written in Javascript) as opposed to the server. It receives the request from the AJAX engine and determines the appropriate webservice to invoke. The webservice returns an XML to which an XSL is applied and the result is displayed in the browser. Such an architecture removes the need of complex MVC frameworks and uses a simple controller written in Javascript.

I am not sure if a light weight, client side MVC framework is what will be used in the future, but I do think there are advantages to such a design. I would like to know what people think. Based on the comments I will discuss pros and cons of such a design in the next few posts.

No comments: