Gossip-Based Real-Time Task Scheduling Using Expander Graph

Abstract

In this paper, we consider the scheduling of real-time distributed tasks in large-scale dynamic networks, where node and link failures and message losses occur frequently. We propose a distributed scheduling algorithm using gossip-based approach called GBTS for dynamic and reliable discovery of suitable nodes that can execute the tasks. GBTS takes advantage of the slack times to optimize the gossiping duration, thereby satisfying the end-to-end timing constraints of tasks with a probabilistic guarantee. Although gossip-based protocols are fault tolerant, they incur high message overheads. We propose to use a highly connected, sparse graph called the expander graph to control the communication complexity of our algorithm. Performance analysis shows that GBTS performs better in terms of both time and message complexity.