Service Degradation-Tolerated Online User Allocation in Edge Computing

Edge computing (EC), as an emerging technology, allows app vendors to deliver low-latency services by allocating users to edge servers nearby. Unfortunately, these heterogeneous and resource-constrained edge servers struggle to serve all the co-existing users with ever-growing service requirements, especially during peak periods. Existing approaches for the user allocation problem insisting on fulfilling prescribed service requirements often fail to tackle this issue and consequently result in a tremendous user loss that significantly impairs app vendors' service profit. We observe that a certain degree of violation of users' requirements can help spare resources to serve more users. Meanwhile, such a violation, which is also called service degradation, can be tolerated by users if they receive appropriate compensation. However, abusive degradation may introduce huge compensation, ultimately leading to reduced service profit. As a result, the service degradation and compensation must be carefully considered by app vendors. In this paper, we study the S ervice D egradation-Tolerated O nline U ser A llocation (SD-OUA) problem in EC environment, aiming to maximize app vendors' service profit. We prove the $\mathcal {NP}$ -hardness of the SD-OUA problem. Then, we propose the D egradation-aware U ser A llocation ( DUA ) approach based on problem reformulation and primal-dual optimization to find its solutions in polynomial time. The performance of DUA is theoretically guaranteed and evaluated against three representative approaches via extensive experiments.