Head-of-the-Line priority with Random Assignment

The well known Head-of-the-Line mechanism can lead to overprovisioning , as the con- straint brought by one of the ∞ows imposes the server speed, while the other ∞ows would require a lower level of resources. We propose and analyse a mechanism, which added to the basic HoL, allows an optimal usage of the server capacity. We derive the conditions under which the optimal behavior can be reached, and we discuss several aspects of the dimensioning.