Lec8Perf

# If your data is not linear massage it so that is

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

This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: (p) = a + c/p a: the sequential part, c: the parallelizable part 10 Plotting hyperbolic functions n༆  A simple way to turn T(p) = c/p into a straight line is to plot its reciprocal: y = 1/T(p) = p/c ¨༊  ¨༊  n༆  This is a straight line with slope 1/c. When analyzing parallel performance we scale this to y’=T(1)/T(p). If T(p) is the time it takes to execute a program with p processors, we call this the speedup of the program In the case of T(p) = a + c/p, the speedup is S(p)=T(1)/T(p)= (a+c)/(a+c/p) ¨༊  For a>0 this is not a straight, but a curve that grows and then flattens out to a constant (a+c)/a Plotting Data: Summary n༆  Visually, a straight line conveys the most information. ¨༊  If your data is not linear, massage it so that is linear, then deduce the original function. n༆  If y=f(x) is polynomial: log y is linear with log x y = f ( x ) = a0 + a1 x + + an x n ≈ an x n (asymptotically) log y = log an + nlog x n༆  If y=f(x) is exponential: log y is linear with x y = f ( x ) = ba x log y = log b + x (log a ) n༆  In the case of T(p) = a + c/p, the speedup is...
View Full Document

## This note was uploaded on 02/12/2014 for the course CS 475 taught by Professor Staff during the Fall '08 term at Colorado State.

Ask a homework question - tutors are online