Analysis of deadline assignment methods in distributed real-time systems

The deadline assignment problem arises in distributed systems where each subtask composing a distributed task must receive a local deadline such that the task end-to-end deadline is met. It also arises in multi-hop networks where the maximum sojourn time of a flow on each visited node must be bounded by a local deadline that allows the flow end-to-end deadline to be met. We first formalize the problem and identify the cases where the choice of a deadline assignment method has a strong impact on system performances. We then propose two deadline assignment methods: Fair Laxity Distribution (FLD) and Unfair Laxity Distribution (ULD). Both assign local deadlines to the flow. These deadlines are based on the flow minimum sojourn time that can be guaranteed on each visited node. FLD and ULD differ in the laxity distribution: fair between the visited nodes for FLD, and proportional to the minimum guaranteed sojourn time for ULD. Performances of FLD and ULD are compared with those of classical methods such as fair assignment and assignment proportional to the workload. Moreover, performance evaluation shows that FLD for NP-EDF scheduling and ULD for FIFO scheduling are good approximations of an optimal algorithm in the context of a video-on-demand system. q 2004 Elsevier B.V. All rights reserved.