Douglas Eadline writes “Can you draw a line with a single point? Sure you can, but you can draw the line anywhere and conclude anything. Measuring network performance on Linux clusters suffers from a similar issue. Often single network performance numbers are quoted when discussing clusters. This article looks beyond single measurements and the subsequent wrong conclusions they can invite. Often times real application data contradicts the conclusions based on single point of performance data.”
Link: clustermonkey.net
