Asymptotically Optimal Algorithm for Online Reconfiguration of Edge-Clouds

"Edge-clouds," which are small servers located close to mobile users, have the potential to greatly reduce delay and backhaul traffic of mobile applications by moving cloud services closer to users at the edge. Due to their limited storage capacity, proper configurations of edge-clouds have a significant impact on their performance. This paper proposes a tractable online algorithm that configures edge-clouds dynamically solely based on past system history without any assumptions on the arrival patterns of mobile applications. We evaluate the competitive ratio, which quantifies the worst-case performance in comparison to an optimal offline policy, of our policy. We prove that the competitive ratio of our policy is linear with the capacity of the edge-cloud. Moreover, we also prove that no deterministic online policy can achieve a competitive ratio that is asymptotically better than ours. The utility of our online policy is further evaluated by traces from real-world data centers. These trace-based simulations demonstrate that our policy has better, or similar, performance compared to many intelligent offline policies that have complete knowledge of all future arrivals.