Virtual Service Provisioning for Internet of Things Applications in Mobile Edge Computing

Abstract

The Internet of Things (IoT) paradigm is paving the way for many new emerging technologies, such as smart grid, industry 4.0, connected cars, smart cities, etc. Mobile Edge Computing (MEC) provides promising solutions to reduce service delays for delay-sensitive IoT applications, where cloudlets are co-located with wireless access points in the proximity of IoT devices. Most mobile users have specified Service Function Chain (SFC) requirements, where an SFC is a sequence of Virtual Network Functions (VNFs). Meanwhile, edge intelligence arises to provision real-time deep neural network (DNN) inference services for users. To accelerate the processing of the DNN inference of a request in an MEC network, the DNN inference model usually can be partitioned into two connected parts: one part is processed on the local IoT device of the request; and the other part is processed on a cloudlet (server) in the MEC network. Also, the DNN inference can be further accelerated by allocating multiple threads of the cloudlet in which the request is assigned. In this thesis, we will focus on virtual service provisioning for IoT applications in MEC Environments. Firstly, we consider the user satisfaction problem of using services jointly provided by an MEC network and a remote cloud for delay-sensitive IoT applications, through maximizing the accumulative user satisfaction when different user services have different service delay requirements. A novel metric to measure user satisfaction of using a service is proposed, and efficient approximation and online algorithms for the defined problems under both static and dynamic user service demands are then devised and analyzed. Secondly, we study service provisioning in an MEC network for multi-source IoT applications with SFC requirements with the aim of minimizing service provisioning cost, where each IoT application has multiple data streams from different sources to be uploaded to the MEC network for processing and storage, while each data stream must pass through the network functions of the SFC of the IoT application, prior to reaching its destination. A service provisioning framework for such multi-source IoT applications is proposed, through uploading stream data from multiple IoT sources, VNF instance placement and sharing, in-network aggregation of data streams, and workload balancing among cloudlets. Efficient algorithms for service provisioning of multi-source IoT applications in MEC networks, built upon the proposed framework, are also proposed. Thirdly, we investigate a novel DNN inference throughput maximization problem in an MEC network with the aim to maximize the number of delay-aware DNN service requests admitted, by accelerating each DNN inference through jointly exploring DNN partitioning and inference parallelism. We devise a constant approximation algorithm for the problem under the offline setting, and an online algorithm with a provable competitive ratio for the problem under the online setting, respectively. Fourthly, we address a robust SFC placement problem with the aim to maximize the expected profit collected by the service provider of an MEC network, under the assumption of both computing resource and data rate demand uncertainties. We start with a special case of the problem where the measurement of the expected demanded resources for each request admission is accurate, under which we propose a near-optimal approximation algorithm for the problem by adopting the Markov approximation technique, which can achieve a provable optimality gap. Then, we extend the proposed approach to the problem of concern, for which we show that the proposed algorithm still is applicable, and the solution delivered has a moderate optimality gap with bounded perturbation errors on the profit measurement. Finally, we summarize the thesis work and explore several potential research topics that are based on the studies in this thesis.