LMR, DTA: adaptive communication algorithms for asynchronous real-time distributed systems using token-ring networks

Abstract

We present two adaptive communication algorithms called LMR and DTA, for asynchronous real-time distributed systems that use IEEE 802.5 token-ring networks and FDDI networks, respectively. The objective of the algorithms is to minimize end-to-end missed-deadline ratio of application tasks. While LMR adapts application tasks to workload fluctuations by reprioritizing trans-node application messages, DTA performs adaptation by dynamically changing token holding times of host machines, respectively. We study the performance of the algorithms through benchmark-driven experimental studies. The performance of the algorithms is compared with an adaptive resource allocation algorithm that performs adaptation by dynamically replicating application processes for load sharing. The experimental results indicate that LMR and DTA outperform the process replication algorithm for load patterns that cause communication latencies to grow faster than execution latencies. Furthermore, we observe that LMR and DTA perform as “good” as the process replication algorithm for load patterns that cause execution latencies to grow faster than communication latencies.