Multi-Resource Fair Allocation for Composited Services in Edge Micro-Clouds
Tongyu Guo, Haitao Zhang, Han Chen Huang, Jianli Guo, Chenze He · 2019
Edge micro-clouds can collaboratively provide computing ability to users nearby in the dynamic environments, such as disaster relief and battlefield. Generally, the applications deployed on edge micro-clouds platform may consist of a set of composited services, and different services require different amounts of multiple dimensional resources. So massive service deployment can lead to unbalanced resource utilization on the micro-clouds. On the other hand, the composited services can be placed in different nodes, and even different edge micro-clouds, and the Quality of Service (QoS) of applications cannot be guaranteed. In this paper, we first define two models of edge micro-clouds and composited services to measure the resource utilization balance degree and QoS constrains respectively. Based on the two models, we formulate the problem of maximizing the edge micro-clouds platform resource utilization balance degree while satisfying the application delay constraints. Subsequently, we propose a Deep Reinforcement Learning (DRL) based multi-resource fair allocation algorithm to obtain the resource allocation strategy by making online decision. In our algorithm, we design a shaped reward function to jointly consider the factors of resource utilization balance degree and application delay constrains, and an improved DRL model is used to improve the convergence performance and the accuracy of the obtained strategy. The experimental evaluation demonstrates the proposed algorithm can guarantee the QoS of applications while achieving a better resource utilization balance degree compared with the other representative algorithms.