Congestion-aware delay-guaranteed scheduling and routing with renewal optimization

Abstract

The emergence of time-critical applications imposes great challenges on traditional best-effort data networks. Such applications demand delay-guaranteed services with different granularity. In this paper, we propose a novel fully distributed approach that provides delay-guaranteed services at the packet level within an autonomous system. Specifically, we adopt priority queues with fixed buffer sizes to provide differentiated services and per-hop delay upper bound and explore path diversity in the network to achieve multiplex gain. Network congestion is a major cause of long delay. To address this issue, a virtual queue manager is deployed at each node in the network to exchange their local queue information with neighbors periodically. The exchanged queue information reflects the congestion status of the neighborhood so the nodes can avoid congested routes in making routing decisions. Given the rich control space including routing and queuing decisions, we aim at maximizing the overall network utility. Due to the randomness caused by network dynamics, we transform the utility maximization problem into renewal optimization which is solved at each node. A delay laxity-based reward function and a weighted queue time cost are designed to characterize each decision. To solve the renewal optimization problem, an algorithm named DSROpt is proposed using an iterative approach. Extensive experiments are conducted to verify the performance of the proposed solution using NS-3. Simulation results show that the proposed solution can guarantee packet-level delay while achieving significant performance improvements in goodput and network utility over the state-of-the-art.