Some parametrized dynamic priority policies for 2-class M/G/1 queues: completeness and applications

TL;DR

This paper establishes the completeness of various dynamic priority policies in 2-class M/G/1 queues, enabling optimal control and resource allocation in diverse systems like networks and manufacturing.

Contribution

It proves the completeness and equivalence of four dynamic priority schemes and applies this to characterize optimal policies in multiple real-world domains.

Findings

Proves completeness and equivalence of four dynamic priority schemes.

Characterizes optimal scheduling policies using these schemes.

Simplifies the $c/\rho$ rule and joint pricing/scheduling problems.

Abstract

Completeness of a dynamic priority scheduling scheme is of fundamental importance for the optimal control of queues in areas as diverse as computer communications, communication networks, supply chains and manufacturing systems. Our first main contribution is to identify the mean waiting time completeness as a unifying aspect for four different dynamic priority scheduling schemes by proving their completeness and equivalence in 2-class M/G/1 queue. These dynamic priority schemes are earliest due date based, head of line priority jump, relative priority, and probabilistic priority. In our second main contribution, we characterize the optimal scheduling policies for the case studies in different domains by exploiting the completeness of above dynamic priority schemes. The major theme of second main contribution is resource allocation/optimal control in revenue management problems for contemporary systems such as cloud computing, high-performance computing, etc., where congestion is inherent. Using completeness and theoretically tractable nature of relative priority policy, we study the impact of approximation in a fairly generic data network utility framework. We introduce the notion of min-max fairness in multi-class queues and show that a simple global FCFS policy is min-max fair. Next, we re-derive the celebrated $c/\rho$ rule for 2-class M/G/1 queues by an elegant argument and also simplify a complex joint pricing and scheduling problem for a wider class of scheduling policies.