Resource Management in Cloud Radio Networks

点击翻译按钮获取中文摘要

The explosive growth in creation and consumption of content on mobile devices has led to a massive increase in mobile data traffic in recent years. In order to satisfy growing user demands, mobile network operators are increasingly moving toward denser deployment of base stations. However, deploying a large number of base stations results in significant increases in capital and operational costs of the network. Cloud radio access network (CRAN) is an emerging mobile network architecture in which signal processing functions are moved to a datacenter, turning base stations into simple low-cost remote radio units. Not only the cloud-based architecture reduces the cost and complexity of deploying more base stations, but also allows signal processing functions to be virtualized in software modules that can be dynamically scaled to adapt to varying user demands, improving network scalability and performance. While CRAN is conceptually simple, several technological and intellectual challenges need to be addressed before it can be realized. A key challenge is the efficient and effective management of intertwined radio (e.g., radio frequency and transmit power) and datacenter (e.g., servers and interconnection links) resources. Our research view is that separate management of these resources is not optimal, even if each one is managed based on state-of-the-art techniques. The goal of this Discovery Program is to introduce and study, in a unified way, optimal or close to optimal algorithms for resource management in CRAN. One of the major concepts we pursue in our research, which has direct practical implications, is proactive resource management, where it is guaranteed that our algorithms perform well under dynamic demands without requiring costly and disruptive reconfigurations. We aim to make foundational contributions toward proactive resource management in cloud radio access networks by developing frameworks to study: i) online resource management, when no information about future demands is available, ii) robust resource management, when only partial information about future demands is available, and iii) autonomic resource management, when the optimal resource management algorithm is learnt autonomously. We focus on CRAN, having future mobile technologies in mind, but the basic tools and approaches to be built and researched are relevant to other cloud-based systems as well. The proposed research will produce new algorithms and theoretical frameworks for resource management in cloud-centric mobile networks. It will provide other researchers with an innovative framework to design autonomic resource management algorithms as well as a suite of efficient algorithms based on conventional optimization techniques, whose performance characteristics and trade-offs are well quantified. Canadian mobile operators and cloud service providers will be able to use our results to better inform their decisions when planning new services and applications.