Serial vs. Concurrent Scheduling of Transmission and Processing Tasks in Collaborative Systems

Abstract

In collaboration architectures, a computer must perform both processing and transmission tasks. Intuitively, it seems that these independent tasks should be executed in concurrent threads. We show that when multiple cores are not available to schedule these tasks, a sequential scheme in which the processing (transmission) task is done first tends to optimize feedback (feedthrough) times for most users. The concurrent policy gives feedback and feedthrough times that are in between the ones supported by the sequential policies. However, in comparison to the process-first policy, it can noticeably degrade feedback times, and in comparison to the transmit-first policy, it can noticeably degrade feedthrough times without noticeably improving feedback times. We present definitions, examples, and simulations that explain and compare these three scheduling schemes for centralized and replicated collaboration architectures using both unicast and multicast communication.