This preview shows page 1. Sign up to view the full content.
Unformatted text preview: testing setup does not provide high enough precision for this. The three series indicate the errors under three different loading conditions. Observe that in all three cases the measurements for run times shorter than around 7.5 ms were very accurate. Thus, our scheme can be 9.4. MEASURING PROGRAM EXECUTION TIME WITH CYCLE COUNTERS 469 Intel Pentium III, Linux
100 10 Measured:Expected Error 1 0.1 Load 1 Load 2 Load 11 0.01 0.001 0 10 20 30 40 50 Expected CPU Time (ms) Figure 9.14: Experimental Validation if Ã -Best Measurement Scheme on Linux System We can consistently obtain very accurate measurements (around 0.1% error) for execution times up to around 8 ms. Beyond this, we encounter a systematic overestimate of around 4 to 6% on a lightly loaded machine and very poor results on a heavily loaded machine. 470 CHAPTER 9. MEASURING PROGRAM EXECUTION TIME used to measure relatively short execution times even on a heavily loaded machine. Series “Load 1” indicates the case where there is only one active process. F...
View Full Document
- Spring '10
- The American