This preview shows page 1. Sign up to view the full content.
Unformatted text preview: G PROGRAM EXECUTION TIME WITH CYCLE COUNTERS
100 475 10 Measured:Expected Error 1 Load 1 Load 2 Load 11 0.1 0.01 0.001 0 10 20 30 40 50 Expected CPU Time (ms) Figure 9.19: Experimental Validation if Ã -Best Measurement Scheme on Compaq Alpha System. For a lightly loaded system, we can consistently obtain accurate ( 1.0% error) measurements. For a heavily loaded system, durations beyond around 10 ms cannot be measured accurately. measurements. In this case, our accuracies were around 0.01 (i.e., 1.0%), rather than 0.001. Still, this is good enough for most applications. In addition, our threshold between reliable and unreliable measurements on a heavily loaded machine was around 48 ms. One interesting feature is that we were sometimes able to get accurate measurements on a heavily loaded machine even for computations ranging up to 245 ms. Evidently, the NT scheduler will sometimes allow processes to remain active for longer durations, but we cannot rely on this property. The Compaq Alpha results are shown in Figure 9.19. Again, we ﬁ...
View Full Document
This note was uploaded on 09/02/2010 for the course ELECTRICAL 360 taught by Professor Schultz during the Spring '10 term at BYU.
- Spring '10
- The American