324_Book

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: ugh for measuring execution times longer than around 7 ms, especially in the presence of other active processes. Unfortunately, we found that our measurement program could not reliably determine whether or not it had obtained an accurate measurement. Our measurement procedure computes a prediction of its error as ´Ú Ú½ µ Ú½ , where Ú is the th smallest measurement. That is, it computes how well it achieves Ô ´Ö µ our convergence criterion. We found these estimates to be wildly optimistic. Even for the Load 11 case, where the measurements were off by a factor of 10, the program consistently estimated its error to be less than ¼ ¼¼½. Setting the value of à In our earlier experiments, we arbitrarily chose a value of 3 for the parameter à , determining the number of measurements we require to be within a small factor of the fastest in order to terminate. To more carefully evaluate the effect of this factor, we performed a series of measurements using values of à ranging from 1 to 5, as shown in Figure 9.15. We performed these measurements for execution times ranging up to 9 ms, since this is the...
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.

Ask a homework question - tutors are online