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: when running on a lightly loaded machine. We exploit these properties to get reliable estimates of true run times. As our object to measure, we used a procedure that repeatedly writes values to an array of 2,048 integers and then reads them back, similar to the code for clear cache. By setting the number of repetitions Ö , we could create computations requiring a range of times. We first determined the expected run time of this procedure as a function of Ö , denoted Ì ´Ö µ, by timing it for Ö ranging from 1 to 10 (giving times ranging from 0.09 to 0.9 milliseconds), and performing a least squares fit to find a formula of the form Ì ´Ö µ ÑÖ · . By using small values of Ö , performing 100 measurements for each value of Ö , and running on a lightly loaded system we were able to get a very accurate characterization of Ì ´Ö µ. Our least squares analysis indicated that the formula Ì ´Ö µ ¾ ¿ Ö ·½ (in units of clock cycles) fits this data with a maximum error less than 0.04%. This gave us confidence...
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