Performability analysis of adaptive drone computation offloading with fog computing

Performance and availability are mutually-dependent quality aspects of drone application systems. Computations on the drone, like image processing, need good performance but may harm the availability of the service. The availability concern can be mitigated by offloading computation tasks to a computing infrastructure like fog computing. However, the latency and the reliability of the communication network have non-negligible impacts on the performance when offloading. This paper focuses on analyzing the tradeoffs between performance and availability in a drone system with various computing modes. We also present a novel adaptive computation offloading scheme called Performability-Aware offloading (PA-Offload). PA-Offload adaptively determines a condition to start or stop offloading by evaluating the expected performance and availability under given environmental status. We introduce a performability measure as an indicator to determine the condition and present comprehensive Stochastic Reward Nets (SRNs) to estimate the expected performabilities for a drone processing system. The numerical studies considering multiple network channels, drones, and fog-computing resources show that PA-Offload can effectively choose better computation modes under different workloads and network conditions. The evaluation results show that PA-Offload consistently achieves the highest performability over the drone processing mode, the fog offloading mode, and other baseline algorithms for offloading decisions. (c) 2023 Elsevier B.V. All rights reserved.