Online Utilization Maximization in Resource Allocation with Minimum Service Guarantees

The natural objective of resource allocation algorithms is twofold: On one hand, to maximize utilization and on the other hand to allow a fair share to all users. The actual meaning of "fair" in this context is manifold; we propose to address fairness in a simple and natural way by guaranteeing a minimum level of service to every user. We develop new competitive online algorithms for this new resource allocation with mandatory service problem and analyze their performance guarantees both in the adversarial-order and random-order online models. We also show that having prior knowledge about the request distribution can be beneficial. We accomplish this by analyzing a probabilistic relaxation of the mandatory service criterion. We study the practical implementation of these theoretical algorithms in the context of online cell selection in access networks. In this setting, mobile users request service and the network needs to assign a relevant cell (or cells) to provide it. We conduct extensive simulations to evaluate the performance of our algorithms in realistic conditions. The results suggest that our new algorithms perform better than applicable adaptations of the commonly used heuristics.