Size: 1083
Comment:
|
Size: 1224
Comment:
|
Deletions are marked like this. | Additions are marked like this. |
Line 29: | Line 29: |
'''We note''' that the geometric mean violates our fundamental principle of performance measurement - it does not predict execution time. |
Back to ComputerTerms
Normalize: Given a reference execution time A, take the execution time you have (B) and divide it by the reference execution time. What you get is a normalized execution time of B with respect to A.
Excution time of B Normalized(B) = ---------------------------- Reference Execution Time A
Given several programs P1, P2, ..., Pn, the average (GeometricMean) normalized execution time is
For (i=1; i++, <= n) { P *= Normalized(Pi) } return P^(1/n) or _____________ / n n / __ / || Normalized(Pi) \/ i=1
Using the geometric mean gives a consistent ranking no matter which machine you normalize to, as in: Computer A is faster than computer B. The Arithmetic mean (average) does not give a consistent ranking when we normalize to different machines. However, the arithmetic mean does give you an execution time average while the geometric mean does not! Instead, it gives you a composite measure.
We note that the geometric mean violates our fundamental principle of performance measurement - it does not predict execution time.
Back to ComputerTerms