2013-05-30: Probabilistic simulation plan with more realistic distribution

May 30, 2013 - We adapt the first study by analysing the effect of σd on the relative ... p
383KB taille 0 téléchargements 266 vues
Probabilistic Simulation Plan with more Realistic Distribution Louis-Claude Canon May 30, 2013∗ The objective is to adapt the previous study to Gamma et Pareto distributions that may be more realistic (the later being heavy tail) and more simple (only two parameters for central tendency, µd and µc , and statistical dispersion, σd and σc ). On June 3, all the XP were relaunched with a correct computation of the gamma distribution parameters.

1

Simulation code

source("function_fix.R")

2

Effect of the communication dispersion

We adapt the first study by analysing the effect of σd on the relative performance of binomial and fibonacci over the dynamic one that is expected to be the best (on average).

2.1

Gamma distribution

The first distribution that is more realistic may be the gamma distribution. It is mentioned in Lublin and Feitelson, 2003 for modeling rigid job runtimes1 . It is also mentioned in Feitelson’s book on workload modeling2 . set.seed(0) n