The experimental setup must control some sources 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: in our ability to accurately predict the actual computation time for the procedure as a function of Ö . We then measured performance using the à -best scheme with parameters à ¿, ¯ ¼ ¼¼½, and Å ¿¼. We did this for a number of values of Ö to get expected run times in a range from 0.27 to 50 milliseconds. For each of the resulting measurements Å ´Ö µ we computed the measurement error Ñ ´Ö µ as Ñ ´Ö µ ´Å ´Ö µ Ì ´Ö µµ Ì ´Ö µ. Figure 9.14 shows an experimental validation of the à -best scheme on an Intel Pentium III running Linux. In this figure we show the measurement error Ñ ´Ö µ as a function of Ì ´Ö µ, where we show Ì ´Ö µ in units of milliseconds. Note that we show Ñ ´Ö µ on a logarithmic scale; each horizontal line represents an order of magnitude difference in measurement error. In order to be accurate within 1% we must have an error below ¼ ¼½. We do not attempt to show any errors smaller than ¼ ¼¼½ (i.e., 0.1%), since our...
View Full Document

Ask a homework question - tutors are online