Đang tải... (xem toàn văn)
User Delay Tolerance Aware Edge Node Placement Optimization for Cost Minimization A thesis submitted in fulfillment of the requirements for the degree of Master of Science Xiaoyu Zhang School of Compu[.]
User-Delay-Tolerance-Aware Edge Node Placement Optimization for Cost Minimization A thesis submitted in fulfillment of the requirements for the degree of Master of Science Xiaoyu Zhang School of Computing Technologies College of Science, Technology, Engineering and Maths RMIT University Melbourne, Victoria, Australia May, 2022 Declaration I certify that except where due acknowledgement has been made, this research is that of the author alone; the content of this research submission is the result of work which has been carried out since the official commencement date of the approved research program; any editorial work, paid or unpaid, carried out by a third party is acknowledged; and, ethics procedures and guidelines have been followed In addition, I certify that this submission contains no material previously submitted for award of any qualification at any other university or institution, unless approved for a jointaward with another institution, and acknowledge that no part of this work will, in the future, be used in a submission in my name, for any other qualification in any university or other tertiary institution without the prior approval of the University, and where applicable, any partner institution responsible for the joint-award of this degree I acknowledge that copyright of any published works contained within this thesis resides with the copyright holder(s) of those works I give permission for the digital version of my research submission to be made available on the web, via the University’s digital research repository, unless permission has been granted by the University to restrict access for a period of time I acknowledge the support I have received for my research through the provision of an Australian Government Research Training Program Scholarship Xiaoyu Zhang School of Computing Technologies, RMIT University 07 March, 2022 Acknowledgement I would like to express my gratitude to my supervisors, Prof Zhifeng Bao and Dr Hai Dong, who guided me throughout this Master by Research study, for their excellent supervision, great engagement, and remarkable support in the last year They offered me a precious and splendid research journey I would also like to thank my family and friends who always support my study I wish to acknowledge the help provided by the staff in the School of Computing Technology of RMIT University Contents Declaration ii Acknowledgement iii Contents iv List of Figures vii List of Tables viii Abstract 1 Introduction 1.1 Motivation 1.2 Research Questions 1.3 Research Contributions 1.4 Thesis Organization 10 Background 11 2.1 MEC Network and Key Components 11 2.2 Literature Review 12 2.2.1 Delay Aware Edge Nodes Placement 12 2.2.2 Cost and Delay Aware Edge Node Placement 14 Problem Formulation 16 iv v CONTENTS 3.1 Preliminaries 16 3.2 Coarse-grained Formulation 18 3.2.1 Workload Measurement 18 3.2.2 Delay Measurement 19 3.2.3 Qualified EN Placement Plan 20 Fine-Grained Formulation 20 3.3.1 Workload Measurement 21 3.3.2 Delay Measurement 21 3.3.3 Qualified EN Placement Plan 22 Cost Minimization in MEC Edge Node Placement Problem 23 3.3 3.4 Hardness Analysis 24 Solutions 26 5.1 5.2 MIP-based Methods 26 5.1.1 Mixed-Integer Programming (MIP) 26 5.1.2 Cluster-based MIP 27 Heuristic-based Methods 30 5.2.1 Coverage First Search (CFS) 30 5.2.2 Distance Aware Coverage First Search (DA-CFS) 31 5.2.3 Fine-grained Optimization 32 Experiments 6.1 6.2 35 Experiment Settings 36 6.1.1 Dataset 36 6.1.2 Experiment Environment 36 6.1.3 Methods for Comparison 36 6.1.4 Parameter Settings 37 6.1.5 Evaluation Metrics 38 Experimental Results 38 6.2.1 38 Effectiveness Analysis vi CONTENTS 6.3 6.2.2 Efficiency Analysis 40 6.2.3 Discussion on Scalability 42 6.2.4 Transmission and Computation Delay in θ 42 6.2.5 The Impact of θ 44 6.2.6 Peak vs AVG 46 6.2.7 Clustering Results for MIP+Cluster 47 6.2.8 Ablation Study on τ for DA-CFS 48 Summary 49 Conclusion 50 Appendices 51 Credits 51 Bibliography 52 List of Figures 1.1 Mobile Cloud Computing Network Architecture 1.2 Mobile Edge (Multi-Access) Computing Network Architecture 1.3 Example of Optimal EN Deployment under Different Delay Tolerance 6.1 Effectiveness and efficiency with different BS input scale 39 6.2 Cost generated by MIP+Cluster with running time limit of 10min and 1h per cluster 41 6.3 Distribution of BSs with different transmission delay proportions on θ 43 6.4 Effectiveness and efficiency with different θ value 45 6.5 Deployment cost: Peak vs AVG 46 6.6 Visualise clusters on Shanghai Telecom Dataset 47 vii List of Tables 1.1 Example of fine-grained workload in BS and EN 2.1 Literature review of delay minimization problems 13 2.2 Literature review of cost minimization problems 14 3.1 Table of notations 17 6.1 Parameter setting 37 6.2 Estimated EN coverage with different θ value 37 6.3 Cluster results on different BS input scale 47 6.4 Cost and # of EN selected with different τ in DA-CFS 48 viii Abstract With the prosperity of 5G technologies, Mobile (or Multi-Access) edge computing (MEC) has emerged as the next generation of network architecture Compared with the previous Mobile Cloud Computing (MCC) architecture, MEC’s decentralising network architecture empowered it amplify user accessible resource and low-latency access Investigating how to effectively place edge nodes, which are also known as edge servers, has become a trendy research topic and attracted wide attention among researchers Among existing studies in this research field, most of the works aim to minimize delays experienced by mobile users, which can be achieved by reaching a workload balance between edge nodes, such that the computation capacity from edge nodes can be fully used However, considering the real use cases, cost-saving, rather than the extremely low network latency, is the prime goal that a service provider (i.e the telecom companies) pursue when deploying edge nodes to a network as users have their delay tolerance within a threshold rather than pursuing a zero-latency network Therefore, how to place edge nodes with the minimum cost, including the construction cost for setting up the edge nodes and the server cost for adding server units to the sites, is still worth further study Although extensive studies have been conducted to find the optimal edge node deployment strategies with the objective of cost minimization, the formulation of such a problem is not close enough to the real-world cases First, they barely take into account the relationship between the allocation of corresponding computation resources and users’ delay tolerance To be more specific, instead of achieving a workload balance on each homogeneous capacitated edge node, the edge nodes can have the heterogeneous capacity, depending on the number of users they would serve and the corresponding workload Second, the delay has not been addressed precisely enough, as most of the existing work only considers the transmission delay while ignoring the computation delay which is decided by the edge node capacity and the workload together In our work, we define a Cost Minimization in MEC Edge Node Placement (CMMENP) problem to find an optimal edge node deployment strategy such that the cost can be minimized without exceeding users’ delay tolerance We precisely define the delay with both transmission delay and computation delay We prove the NP-hardness of this problem and propose a series of solutions, including Clusterbased Mixed Integer Programming, Coverage First Search, and Distance-Aware Coverage First Search Intuitively, the Cluster-based Mixed Integer Programming solves the efficiency issue in the pure MIP solution by sacrificing some degree of accuracy, while the Coverage First Search is an efficiency prioritized solution with Distance-Aware Coverage First Search to further improve its effectiveness without loss of efficiency In addition, we propose a fine-grained optimization strategy to delicately allocate computation resources to edge nodes at the user service request level, which can remarkably decrease the deployment cost Comprehensive experiments have been conducted on a large-scale real-world dataset, which demonstrates the competitiveness of our solutions in terms of efficiency, effectiveness and scalability compared with the state-of-art work ... Minimization Prioritized Edge Nodes Placement and Cost Aware Edge Node Placement 2.2.1 Delay Aware Edge Nodes Placement Minimizing delay is a hot research topic in Edge nodes placement As shown in Table... 12 2.2.1 Delay Aware Edge Nodes Placement 12 2.2.2 Cost and Delay Aware Edge Node Placement 14 Problem Formulation 16 iv v CONTENTS 3.1... overall cost With the objective of minimizing the total cost, the optimal edge node deployment strategy would vary under different users’ delay tolerance Fig 1.3b illustrates an optimal edge node placement