11.2. Identify a Problem

As with the performance problem in the preceding chapter, we have to define what we will investigate and try to overcome it. In this case, we will optimize the time to open a pop-up menu when using the nautilus file manager for the GNOME desktop. In nautilus, pop-up menus are opened by right-clicking anywhere in a nautilus file management window. In this particular case, we will be investigating the performance of the pop-up menus that appear when we right-click the background of an open window rather than when we right-click a particular file or folder.

Why should we optimize this? Even though the amount of time to open a pop-up may be less than a second, it is still slow enough that users can perceive the lag between when they right-click the mouse and when the menu shows up. This sluggish pop-up gives the GNOME user the impression that the computer is running slowly. People notice a slight delay, and it can make interaction with nautilus annoying or give the impression that the desktop is slow.

This particular performance problem is different from the GIMP problem of the preceding chapter. First, the core components of the desktop (in this case, GNOME) are typically more complicated and interlocked than a typical desktop application. The components typically rely on a variety of subsystems and shared libraries to do their work. Whereas the GIMP was a relatively self-contained application, making it easier to profile and recompile when necessary, the GNOME desktop is made up of many different interlocking components. The components may require multiple processes and shared libraries, each performing a different task on behalf of the desktop. nautilus, in particular, is linked to 72 different shared libraries. Tracking down exactly which piece of code is spending time, how much it is spending, and why it is spending it, can be a daunting task.

The significant second difference of this performance investigation from the GIMP investigation is that the times we are trying to reduce are on the order of milliseconds rather than seconds or minutes. When the times are so small, it can be difficult to make sure that the profiling data that you are capturing is actually the result of the event that you are trying to measure rather than just the noise around trying to stop and start the profiling tools. However, this short time period also makes it practical to trace all aspects of what the application does for the interesting period of time.