This preview shows page 1. Sign up to view the full content.
Unformatted text preview: or execution times above 10 ms, the measurements ÌÑ consistently overestimate the computation times Ì by around 4 to 6%. These overestimates are due to the time spent handling timer interrupts. They are consistent with the trace shown in Figure 9.3, showing that even on a lightly loaded machine, an application program can execute for only 95 to 96% of the time. Series “Load 2” and “Load 11” show the performance when other processes are actively executing. In both cases, the measurements become hopelessly inaccurate for execution times above around 7 ms. Note that an error of 1.0 means that ÌÑ is twice Ì , while an error of 10.0 means that ÌÑ is eleven times greater than Ì . Evidently, the operating system schedules each active process for one time interval. When Ò processes are active, each one only gets ½ Òth of the processor time. From these results, we conclude that the Ã -best scheme provides accurate results only for very short computations. It is not really good eno...
View Full Document
- Spring '10
- The American