Enhanced WCTT Analysis of Distributed Networked Systems with Cyclic Dependencies

A relevant issue for distributed networked systems is handling cyclic dependencies at the network level to prove the guaranteed Worst-Case Traversal Times (WCTT). To deal with the performance evaluation of such systems, most relevant existing techniques are based on local analyses of delay upper bounds in each crossed node, resulting in pessimistic end-to-end delay bounds. To overcome this limitation, an enhanced WCTT global analysis is proposed in this paper to improve the delay bounds tightness, while handling large-scale systems. The main idea consists in extending the most recent concept in Network Calculus, i.e. Pay Multiplex Only Once (PMOO), to networks with cyclic dependencies. The PMOO principle consists in paying the bursts of interfering flows only once, when accounting the flow serialization phenomena along the flow path. Extensive sensitivity analysis of the proposed approach, when varying the network size and utilization rate, has been conducted regarding the system performance, in terms of system scalability and resource-efficiency. The results of this evaluation clearly show that our approach outperforms the conventional WCTT analyses, leading to less pessimistic delay bounds for large-scale and highly loaded networked systems, i.e., guaranteed stability for a system configuration of $100$ nodes under full utilization rate.