HAHS: a fault-tolerant hyper-allocate & hyper-switch framework for fog computing
Abstract
Abstract Fog and edge computing platforms must ensure real-time QoS (quality of service) despite frequent VM (virtual machine) failures and highly dynamic workloads. To address this challenge, we propose the HAHS (Hyper-Allocate & Hyper-Switch) framework, a fault-tolerant scheduling solution that integrates two complementary strategies: Hyper-Allocate, which performs multi-criteria VM scoring, and Hyper-Switch, which leverages UCB1 (Upper Confidence Bound) with an $$\epsilon$$ -greedy policy to adaptively select among meta-heuristics under a strict sub-millisecond decision budget. HAHS is implemented on an enhanced version of with multi-level failure injection, DAG (directed acyclic graph)-aware workloads, and realistic system overheads. Extensive experiments across diverse fault rates (0–80%), scales (20–200 VMs), and task volumes (up to 100,000) demonstrate that HAHS consistently outperforms competing approaches. Under extreme conditions with 80% VM failures, HAHS improves task success rates by 10–15% compared with the second-best method, while maintaining low completion-time inflation. Moreover, HAHS exhibits robust scalability, stable performance rankings across workload intensities, and the smallest performance degradation when transitioning from fault-free to high-fault scenarios. These results highlight HAHS as a practical, resilient, and efficient scheduler for fault-prone fog and edge environments.