Our goal in this section is to evaluate the effectiveness of PAVM when running real-world applications in a working system, and see how it compares to some other power-management policies. All workloads are executed on a Pentium 4 PC, with 512 MB of RDRAM (16 devices), running Linux 2.4.18 kernel. We define three types of workloads: Light, Poweruser, and Multimedia. These workloads are composed of different sets of user applications as shown in Table 5. The applications are not meant to be comprehensive, but rather found to be representative of the type of workloads used most often by us and our colleagues on mobile platforms.
Our representative Light workload consists of web browsing, with some e-mailing and word processing, while listening to mp3 music in the background, all run in a windowed, graphical environment. This type of workload is most commonly used on mobile platforms, and the workload is characterized as mostly idle. However, some users (Powerusers), due to the nature of their work (e.g., graphics designing, programming), utilize and stress their systems more. Their workloads can be characterized by repeated periods of low system utilization (designing, coding) followed by periods of high system utilization (rendering, compiling). To simulate this type of workload, we add periodic Linux kernel compilations to generate periods of heavy load on top of the Light workload. With a growing number of multimedia-rich applications, users may impose even heavier workloads on their mobile systems (e.g., 3D gaming, playing video). Multimedia workloads keep the system in a high utilization state continuously for a long period of time. To simulate this and keep workloads consistent across different experiments, we play an MPEG4-encoded movie using the Xine video player in full-screen mode.