R-PF: Enhancing Service Regularity for Legacy Scheduling Policy

Abstract

We present a novel scheduling policy, regularity-aware proportional fair (R-PF) that enhances service regularity for the legacy proportional fair (PF) scheduling policy. R-PF addresses the challenge that clients may have different preferences between service regularity and long-term average throughput. R-PF allows clients to specify their own preferences, and then provides services tailored to clients’ specifications. R-PF preserves the advantages of PF, including high spectrum efficiency, fairness among clients, minimal coordination between the base station and clients, and low complexity. We analytically study the performance of R-PF under an i.i.d. channel model. We prove that each client achieves its optimal tradeoff between service regularity and throughput by reporting its true preference. We also prove that the total throughput under R-PF is almost the same as that under PF. Furthermore, we compare the performance of R-PF against other state-of-the-art scheduling policies using trace-based simulations. Simulation results demonstrate that our policy provides the optimal performance for all kinds of clients with different preferences between service regularity and throughput. Simulation results also show that R-PF is backward compatible in that it enhances service regularity for clients who prefer high regularity without sacrificing the throughput for those who prefer high throughput.