A solution for the priority queue problem of deadline-ordered service disciplines

Abstract

To provide service guarantees in packet-switching networks, several deadline-ordered service disciplines have been proposed, including delay-EDD (earliest due deadline), jitter-EDD, leave-in-time, RFS, and virtual-clock. In a deadline-ordered service discipline, packets are assigned transmission deadlines and are transmitted according to the deadline-ordered scheduling policy: transmit packets in increasing order of deadlines. Although these service disciplines provide desirable quality of service (QoS) guarantees, they have been regarded as infeasible for high-speed networks due to the high overhead imposed by the sorting of packets in the server queue. This paper shows that these service disciplines become efficient if they replace the deadline-ordered scheduling policy with a new scheduling policy that is defined. This new scheduling policy uses hardware support to achieve an O(1) time complexity. We prove that a server that employs this new scheduling policy provides a delay bound to real-time packets that is slightly larger than what the server can provide with deadline-ordered scheduling.