This preview shows page 1. Sign up to view the full content.
Unformatted text preview: upper limit of times for which our scheme can get useful results. When we have Ã ½, the procedure returns after making a single measurement. This can yield highly erratic results, especially when the machine is heavily loaded. If a timer interrupt happens to occur, the result is extremely inaccurate. Even without such a catastrophic event, the measurements will be subject to many sources of inaccuracy. Setting Ã to 2 greatly improves the accuracy. For execution times less than 5 ms, we consistently get accuracy better than 0.1%. Setting Ã even higher gives better results, both in consistency and accuracy, up to a limit of around 8 ms. These experiments show that our initial guess of Ã ¿ is a reasonable choice. Compensating for Timer Interrupt Handling The timer interrupts occur in a predictable way and cause a large systematic error in our measurements for execution times over around 7 ms. It would be good to remove this bias by subtracting from the measured run time for a program an estimate of t...
View Full Document
- Spring '10
- The American