Control of multi-resource infrastructures in mobile cloud-edge computing

Abstract

Today's infrastructures have changed more complex than in the past. Various types of nodes with heterogeneous processing, sensing and storage resources are highly connected by networking resources including wired and wireless links. As a result, several service types assisted by multi-resource types have emerged, such as content caching service, computation offloading service, and network function virtualization service. For these services types, an efficient method controlling multi-resource infrastructures is needed because the quality of the services (QoS) cannot be guaranteed when only one of the required resource is not sufficient. In this paper, we study how to efficiently manage multi-resource types under several scenarios in order to enhance the quality of services, reduce system cost (e.g., delay, energy, financial cost), and support service requests as much as possible.

First, we consider content caching service assisted by networking and storage resources. We study a hybrid caching problem for achieving throughput optimality in cloud and edge-based 5G wireless networks. We propose dynamic caching algorithms considering content request arrival from mobile users, traffic congestion at backhaul, fronthaul and radio access networks, heterogeneous content sizes, and different storage capacity of cloud and edge nodes. Our caching decisions can be calculated in polynomial time while keeping constant approximation ratios to the throughput optimality.

Next, we consider mobile computation offloading service assisted by networking and processing resources. First, we study a computation offloading problem in a mobile device for achieving application throughput fairness and energy efficiency. We propose a dynamic offloading policy which controls throughput of each application, scheduling for local processing and computation offloading, CPU clock scaling, and network interface selection. Second, we study a data center management problem for achieving a maximum profit of a computation offloading service provider. We propose time-dependent pricing and dynamic server provisioning algorithm. We prove that our algorithms achieve optimal performance in a long-term sense.

Finally, we consider a general framework for infrastructures and services, assisted by multi-resource types. First, we study a multi-resource management problem when a traditional network service, network function virtualization service and computation offloading service coexist. We propose a extragradient-based sending rate control and multi-path routing algorithm in order to maximize the quality of services and while reducing the incurred system costs. We show that the proposed algorithm can be decentralized, and prove that it converges to the optimal solution. Second, we study a multi-resource management problem for Radio-Cloud cellular network systems where the core and edge resources are connected in a hierarchical structure. We propose a traffic arrival control, dynamic service chaining, process/network scheduling and beam on/off & user scheduling algorithm in order to maximize the user satisfaction.