Throughput Optimal Decentralized Scheduling of Multi-Hop Networks with End-to-End Deadline Constraints: Unreliable Links

TL;DR

This paper presents a novel distributed scheduling policy for unreliable multi-hop networks with end-to-end deadlines, optimizing throughput under power constraints without requiring network state knowledge.

Contribution

It introduces a new stochastic decomposition approach to derive an optimal distributed scheduling policy for multi-hop networks with deadline constraints.

Findings

Achieves optimal throughput vector under average-power constraints.

Decentralized decisions based solely on packet age, no network state needed.

Provides near-optimal policy with peak-power constraints.

Abstract

We consider unreliable multi-hop networks serving multiple flows in which packets not delivered to their destination nodes by their deadlines are dropped. We address the design of policies for routing and scheduling packets that optimize any specified weighted average of the throughputs of the flows. We provide a new approach which directly yields an optimal distributed scheduling policy that attains any desired maximal timely-throughput vector under average-power constraints on the nodes. It pursues a novel intrinsically stochastic decomposition of the Lagrangian of the constrained network-wide MDP rather than of the fluid model. All decisions regarding a packet's transmission scheduling, transmit power level, and routing, are completely distributed, based solely on the age of the packet, not requiring any knowledge of network state or queue lengths at any of the nodes. Global coordination is achieved through a tractably computable "price" for transmission energy. This price is different from that used to derive the backpressure policy where price corresponds to queue lengths. A quantifiably near-optimal policy is provided if nodes have peak-power constraints.