Dynamic service placement and request scheduling for edge networks

Abstract

Multi-access computing (MEC) represents an effective technology to satisfy low-latency demands by capacitating the execution of data-intensive and latency-sensitive tasks at the edge of mobile networks. While recent studies have investigated cost-awareness regarding where to execute service tasks and how to schedule user requests to edge servers, most have primarily leveraged deep reinforcement learning or concentrated on less-complicated assumptions. In this paper, we design an online algorithm for joint service placement and request scheduling in MEC networks, subjected to multi-dimensional constraints, aiming to minimize the operational cost. However, the problem is non-trivial since it involves time-correlated service placement cost and future information. Our proposed online algorithm employs the online learning technique and rounding method to address these challenges, consisting of the following basic modules: (1) a regularization approach to decouple the offline cost-minimization problem into multiple convex sub-problems, each to be efficiently solved in each time slot; (2) a rounding method to transform an optimal fractional solution of the convex sub-problem into an integer solution of the original cost-minimization problem, with provable the performance guarantee. Our analytical results and simulations verify the effectiveness of the proposed online algorithm.