Determining How the Application is Using System Resources

Choosing an Effective Benchmark

To accurately verify whether or not your application is reaching maximum performance levels, you will need to run some benchmarks. An effective benchmark focuses on the critical functionality of the application. What tasks does a typical end-user repeat most often while using your application? Is the data used in the application typical of the complexity of the data that your end users work with?

In addition, it is important that the benchmark does not spend a lot of time in startup activities, but does spend enough time on other tasks to give you an accurate understanding of performance issues. Benchmark performance should scale on different systems according to the published specifications. If it doesn't, your benchmark is probably not spending enough time on the critical application tasks, in this case, rendering graphics.

Finally, the benchmark results should be repeatable, with approximately the same timing measurements for each time a given task is run.

Identify the Bottlenecks

If the use of system resources is not balanced by the application, performance can slow down considerably. Performance bottlenecks can occur in many different places in your applications, including: graphics, the CPU, memory, the network, and I/O systems. For example, an application that does extensive mathematical computations on data before displaying that data may be CPU-bound. In other words, it will spend a long time processing data and using the CPU resources before sending any information down for graphics processing, although it might appear as if graphics is slow.

It is critical that you understand how the system resources are being used before beginning to tune your code. Otherwise, the time spent tuning code may have little impact on your overall application performance. For example, if graphics is the bottleneck and you are already getting maximum performance from the graphics hardware, no amount of change to the interactions between the graphics and your application will improve your application's performance. Similarly, if the network is the bottleneck, no amount of graphics tuning will improve application performance.