ebook img

Resource Allocation Optimization in Large Scale Distributed Systems PDF

215 Pages·2014·4.12 MB·English
by  
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Resource Allocation Optimization in Large Scale Distributed Systems

AN ABSTRACT OF THE DISSERTATION OF Thuan Duong-Ba for the degree of Doctor of Philosophy in Electrical and Computer Engineering presented on May 16, 2014. Title: Resource Allocation Optimization in Large Scale Distributed Systems Abstract approved: Thinh P. Nguyen Bella Bose We studied the problem of resource allocation in large scale distributed applications such as Online Social Networks (OSN) and Cloud Computing. In such settings, resource al­ location schemes need to efficient as well as adaptive to the time-varying environments. The abstract resource allocation problem concerns with how to optimally use resources for different tasks. In the context of this dissertation, the resources are servers and the tasks are (a) the virtual machines in the cloud computing setting, and or users for on- line social network applications. It is well-known that the general resource allocation problem is NP-hard. Therefore, in this dissertation, we study a number of heuristic algorithms designed for two primary objectives: 1) achieve reliability via load balanc­ ing among resource providers and 2) minimizing the energy consumption by reducing unnecessary intercommunication loads among the servers. Specifically, the dissertation has three main components. The first component deals with optimal assignment of user data to servers to maximize load balance and mini­ mize power consumption. In this component, we propose a novel Distributed Perturbed Greedy Search (DPGS) algorithm which combine both deterministic search and random search to speed the convergence while avoiding local optimum. The empirical shows that the DPGS has a fast convergence rate to the near optimal solution even when the environment changes. The second component deals with the analysis on the conver­ gence rates of a general simulated annealing algorithm via the notion of adiabatic time. We then apply the results to characterize the convergence rates for simulated anneal­ ing algorithm when applied to the optimal assignment in the component one. Finally, the third component of the dissertation is concerned with optimal assignment of virtual machines to servers in the context of cloud computing, in order to minimize the energy subject to a given performance requirement. We show that the problem can be approx­ imated well as a convex problem, and propose convex relaxation technique to find the optimal solution. ©c Copyright by Thuan Duong-Ba May 16, 2014 All Rights Reserved Resource Allocation Optimization in Large Scale Distributed Systems by Thuan Duong-Ba A DISSERTATION submitted to Oregon State University in partial fulfillment of the requirements for the degree of Doctor of Philosophy Presented May 16, 2014 Commencement June 2014 Doctor of Philosophy dissertation of Thuan Duong-Ba presented on May 16, 2014. APPROVED: Major Professor, representing Electrical and Computer Engineering Director of the School of Electrical Engineering and Computer Science Dean of the Graduate School I understand that my dissertation will become part of the permanent collection of Oregon State University libraries. My signature below authorizes release of my dissertation to any reader upon request. Thuan Duong-Ba, Author ACKNOWLEDGEMENTS First and foremost, I would like to express my acknowledgment of and deepest grat­ itude to my advisors, Professor Thinh Nguyen and Professor Bella Bose, for giving me the opportunity to become a Ph.D. student and for their insight, great guidance, and en­ couragement during my studies at Oregon State University. Without their support this dissertation would not have been possible. During my research, I have also had opportunities of cooperating with many great people at the School of Electrical Engineering and Computer Science. Professor Patrick Chiang, thank you very much for the fruitful cooperation and guidance during my first project. I would like to thank Professor Bechir Hamdaoui and Professor Yevgeniy Kovchegov for their acceptance to be my committee members as well as valuable com­ ments on my dissertation. I am also indebted to all my friends for the nice time we have spent together. Es­ pecially lab-mates who helped me so much from the first day I joined the group. They helped me not only in getting familiar with a totally new life in the US, but also in research. Last but not least, I would like to express my sincere appreciation to my family and relatives for continuously supporting and encouraging me during my studying in the US. Finally, I want to thank my wife Huong Nguyen and my two children Tung and Nhung for providing me a happy family with constant support and encouragement. TABLE OF CONTENTS Page 1 Introduction 1 1.1 Resource Allocation in Distributed Systems . . . . . . . . . . . . . . . . 1 1.1.1 OptimizationFormulation . . . . . . . . . . . . . . . . . . . . 3 1.1.2 Multi-objectiveOptimization . . . . . . . . . . . . . . . . . . . 4 1.1.3 An Example of Resource Allocation Problems . . . . . . . . . . 6 1.2 ContributionofThisDissertation . . . . . . . . . . . . . . . . . . . . . . 7 1.2.1 Resource Allocation for Load Balancing . . . . . . . . . . . . . 7 1.2.2 Resource Allocation for Energy Conservation . . . . . . . . . . 7 1.2.3 Convergence Rate of Simulated Annealing Algorithms . . . . . 8 1.3 OrganizationofThisDissertation . . . . . . . . . . . . . . . . . . . . . . 8 2 Background 9 2.1 OnlineSocialNetworkApplications . . . . . . . . . . . . . . . . . . . . 9 2.1.1 SocialNetworkAnalysis . . . . . . . . . . . . . . . . . . . . . 10 2.1.2 NetworkModels . . . . . . . . . . . . . . . . . . . . . . . . . 12 2.1.3 NetworkGrowthModels . . . . . . . . . . . . . . . . . . . . . 13 2.2 GraphPartitioningProblem . . . . . . . . . . . . . . . . . . . . . . . . . 14 2.2.1 LocalSearchMethods . . . . . . . . . . . . . . . . . . . . . . 15 2.2.2 SpectralGraphPartitioning . . . . . . . . . . . . . . . . . . . . 18 2.2.3 Multi-levelGraphPartitioning . . . . . . . . . . . . . . . . . . 19 2.3 CloudComputing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 2.3.1 TypesofServices . . . . . . . . . . . . . . . . . . . . . . . . . 21 2.3.2 TypesofCloudComputing . . . . . . . . . . . . . . . . . . . . 22 2.3.3 AttributesofCloudComputing . . . . . . . . . . . . . . . . . . 23 2.3.4 ResourceAllocation . . . . . . . . . . . . . . . . . . . . . . . . 24 2.4 Bin-PackingProblem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 2.4.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 2.4.2 Hardness of The Bin-Packing Problem . . . . . . . . . . . . . . 26 2.4.3 GreedyAlgorithms . . . . . . . . . . . . . . . . . . . . . . . . 27 2.5 MarkovChainMonteCarlo(MCMC)Method . . . . . . . . . . . . . . . 29 2.5.1 MarkovChain . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 2.5.2 ConvergenceRates . . . . . . . . . . . . . . . . . . . . . . . . 33 2.5.3 SimulatedAnnealing . . . . . . . . . . . . . . . . . . . . . . . 36 2.6 ConvexOptimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 2.6.1 ConvexSets . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 TABLE OF CONTENTS (Continued) Page 2.6.2 ConvexFunctions . . . . . . . . . . . . . . . . . . . . . . . . . 39 2.6.3 ConvexOptimizationProblems . . . . . . . . . . . . . . . . . . 42 2.6.4 Duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 2.6.5 Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 3 Distributed Algorithms For Load Balancing In Large Scale Applications 58 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 3.2 RelatedWork . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 3.3 ProblemFormulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 3.3.1 AlgorithmicOutline . . . . . . . . . . . . . . . . . . . . . . . . 72 3.3.2 TechnicalBuildingBlocks . . . . . . . . . . . . . . . . . . . . 73 3.4 Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 3.4.1 Centralized Simulated Annealing Algorithm . . . . . . . . . . . 76 3.4.2 Distributed Data Structures and Computations . . . . . . . . . . 77 3.4.3 Distributed Simulated Algorithm (DSA) . . . . . . . . . . . . . 81 3.4.4 Distributed Perturbed Greedy Search (DPGS) . . . . . . . . . . 82 3.4.5 ConvergenceAnalysisofDPGS . . . . . . . . . . . . . . . . . 86 3.5 SimulationResults . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 3.5.1 SmallSyntheticGraphs . . . . . . . . . . . . . . . . . . . . . . 97 3.5.2 LargeGraphs . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106 4 Distributed Assignment And Replication For Online Social Networking Appli­ cations 107 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 4.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 4.3 ProblemFormulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 4.4 DistributedAlgorithm/Protocol . . . . . . . . . . . . . . . . . . . . . . . 115 4.5 SimulationResults . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 TABLE OF CONTENTS (Continued) Page 5 Convergence Rate of Inhommogeneous Markov Chains 125 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 5.2 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127 5.2.1 SimulatedAnnealingAlgorithm . . . . . . . . . . . . . . . . . 127 5.2.2 MarkovChainProperties . . . . . . . . . . . . . . . . . . . . . 129 5.3 Analysis Framework for Simulated Annealing . . . . . . . . . . . . . . . 131 5.4 Application to Client-Server Assignment . . . . . . . . . . . . . . . . . . 139 5.4.1 ProblemFormulation . . . . . . . . . . . . . . . . . . . . . . . 139 5.4.2 Simulated Annealing Based Algorithm . . . . . . . . . . . . . . 141 5.5 Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154 5.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157 6 Joint Virtual Machine Placement and Migration Scheme for Datacenters 158 6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158 6.2 System Model, Assumptions, and Optimization Objectives . . . . . . . . 161 6.3 Multi-Objective Optimization to VM Management . . . . . . . . . . . . 164 6.3.1 Multi-Objective Optimization Function . . . . . . . . . . . . . 164 6.3.2 Graph Partitioning Based Rank Minimization . . . . . . . . . . 170 6.4 Relaxed Convex Optimization Based Heuristic Algorithm . . . . . . . . . 175 6.4.1 RelaxedConvexOptimization . . . . . . . . . . . . . . . . . . 175 6.4.2 Joint VM Placement and Migration (JPM) Algorithm . . . . . . 176 6.5 PerformanceEvaluationandDiscussion . . . . . . . . . . . . . . . . . . 179 6.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181 7 Conclusion and Future Work 186 7.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186 7.2 FutureWork . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188 Bibliography 189

Description:
algorithms designed for two primary objectives: 1) achieve reliability via load c©Copyright by Thuan Duong-Ba . 2.4.2 Hardness of The Bin-Packing Problem . Pinterest. Figure 1.1: User growth rates of popular Social Networks.
See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.