10.14. Chapter Summary

In this chapter, we determined why an application (the GIMP filter lic) was CPU-bound. We figured out the base runtime of the application, set a goal for optimization, and saved a reference image to verify that our optimizations did not change the behavior of the application. We used the Linux CPU performance tools (oprofile and ltrace) to investigate exactly why the application was CPU-bound. We then used the Web to understand how the application worked and to figure out different ways to optimize it. We tried a few different optimizations, but ultimately, we chose the classic performance trade-off of increased memory usage for reduced CPU usage.

We beat our optimization goal, and then verified that our optimization did not change the output of the application.

Whereas this chapter focused on optimizing a single application's runtime, the next chapter presents a performance hunt that concentrates on reducing the amount of latency when interacting with X Windows. Reducing latency can be tricky, because a single event often sets off a nonobvious set of other events. The hard part is figuring out what events are being called and how long each of them are taking.