Proportional-share scheduling in single-server and multiple-server computing systems

Proportional Share Scheduling (PSS), which is the allocation of prespecified fractions of a certain resource to different classes of customers, has been studied both in the context of the allocation of network bandwidth and of processors. Much of this work has focused on systems with a single scheduler and when all classes of customers are constantly backlogged. We study the objectives and performance of PSS policies for processor scheduling when these conditions do not hold