Use median/absolute deviation instead of mean/stddev

[jonnor]

Currently we are very sensitive to outliers when determining whether we can reallize a particular configuration. For instance service disruption with high processing times for just 1 hour during a week can make scaling configs be deemed unrealizable.
Using median, and the associated median absolute deviation, should make us more robust in these cases.
https://en.wikipedia.org/wiki/Median_absolute_deviation
http://www.r-bloggers.com/absolute-deviation-around-the-median/