Performance Analysis of J2EE Applications Using AOP Techniques

In a complex distributed computing environment like J2EE, it is very difficult to pinpoint the component that is causing a performance bottleneck. Applications can be profiled by including instrumentation code manually, but this could be cumbersome and time-consuming, and might impact the stability of the application itself. Aspect-Oriented Programming (AOP) technology can be elegantly and effectively applied for performance analysis, as illustrated by Davies et al.

Aspect-oriented programming allows the programmer to inject pieces of functionality into existing code. This can be done either during compile time (AspectJ) or during run time (Aspectwerkz). The functionality that is injected typically addresses cross-cutting concerns spread among existing code pieces. In AOP terminology, such functionality that can be injected into existing code is termed an advice. The point of execution in the existing code where the advice needs to be applied is termed a point-cut. The point-cut together with an advice is termed as an aspect. For more information on AOP, refer to Graham O'Regan's ONJava article "Introduction to Aspect-Oriented Programming."

In this article we demonstrate the use of AOP techniques through which J2EE applications can be easily instrumented without any modifications to application code. We have developed a very simple tool to achieve the above objective. Since the instrumentation has very low overhead, this tool can be deployed in the staging environments to identify problematic Java method calls and SQL statements.

We describe the architecture of the profiling tool and then the advices that were developed to instrument the application. This is followed by an illustration of how the instrumentation can be added to the necessary method calls through point-cuts, and finally, we show some of the results obtained through this tool.