Our measurement procedure computes a prediction of

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: ng to keystrokes and other service requests. Observe the wide range of error values shown on this graph. In general, only measurements that are within ¦10% of the true value are acceptable, and hence we want only errors ranging from around ¼ ½ to ·¼ ½. Below around 100 ms (10 timer intervals), the measurements are not at all accurate due to the coarseness of the timing method. Interval counting is only useful for measuring relatively long computations— 100,000,000 clock cycles or more. Beyond this, we see that the error generally ranges between ¼ ¼ and ¼ ½, that is, up to 10% error. There is no noticeable difference between the two different loading conditions. Notice also that the errors have a positive bias; the average error for all measurements with ÌÑ ½¼¼ms is 1.04, due to the fact that the timer interrupts are consuming around 4% of the CPU time. These experiments show that the process timers are useful only for getting approximate values of program performance. They are too coarse-grained to use for any measurement having dur...
View Full Document

Ask a homework question - tutors are online