A dark and stormy night: Reallocation storms in edge computing

Abstract Efficient resource usage in edge computing requires clever allocation of the workload application components. In this paper, we show that under certain circumstances, number superfluous reallocations from one server to another may grow a significant proportion all user tasks—a phenomenon present as reallocation storm. We showcase on city-scale deployment by simulating task workloads scenarios capturing likely deployments and patterns. The simulations are based large real-world data set city-wide Wi-Fi network connections, with more than 47M connections over ca. 560 access points. study occurrence storms three common edge-based strategies compare latency–workload trade-offs related each strategy. As result, find vanish when capacity is increased above threshold, unique for strategy, peaking at 35% peak ES workload. Further, while strategy aiming minimize latency consistently resulted worst storms, two other strategies, namely random bottom-up which always chooses lowest target, behave nearly identically terms well storm dense deployments. Since much less coordination, recommend it Moreover, conditions associated storms. discover servers very highest best around few busy nodes thus mirroring their identify circumstances an elevated risk such summertime (ca. 4 times average) weekends 1.5 risk). Furthermore, mass events popular sports games incurred high (nearly 10 MEC-based scenario.