Reinforcement Learning-enabled Auctions for Self-Healing in Service Function Chaining

Service Function Chaining (SFC), defines the capability of interconnecting a number of ordered Service Functions (SFs) to create composite network services. A critical issue in SFC is the autonomic fault recovery, i.e., bringing the system back to its normal operation after a hardware or software failure. To address this challenge, in this paper, we propose a novel distributed methodology that treats the SFC Self-Healing problem in an Edge-Cloud infrastructure, while accounting for the various stakeholders. In particular, the individual SFC healing decisions are iteratively optimized and determined, while a Reinforcement Learning (RL)-based SFC-to-datacenter association procedure is realized. This process is complemented by a combinatorial auction-based resource allocation mechanism that resolves the potential SFC collocations at the end of each iteration. The proper operation, effectiveness and efficiency of our proposed healing mechanism is assessed under various evaluation scenarios.