Results

The purpose of our study is to determine if the heuristics developed in prior studies can be practically applied to actual pocket computers. We examined a number of policies, most of which are variants of the $\mbox{AVG}_{\mbox{\em N}}$ policy. As described in §4.3, we used three different speed setting policies. Our intent was to focus on systems that could be implemented in an actual O/S and that did not require modifications to the applications (such as requiring information about deadlines or schedules). We assumed that our workloads had inelastic constraints; in other words, we assumed the applications had no way to accommodate ``missed deadlines''.

We split the discussion of our results into three parts. The first section describes aspects of the applications and how they differ from those used in prior work and the second section discusses the performance of the different clock scheduling algorithms. Finally, we examine the benefit of the limited voltage scaling available on the Itsy and summarize the results.