Availability Modeling and Analysis of Data Center Systems using Stochastic Reward Nets : 데이터 센터 시스템의 가용성 모델링 및 분석

다국어 초록 (Multilingual Abstract)

This dissertation aims to develop analytic stochastic models for availability assessment
and prediction of contemporary Data Center System (DCS)s including (i) Virtualized Servers
System (VSS); (ii) Data Center Network (DCN); (iii) Software Defined Network (SDN); and
(iv) Disaster Tolerant Data Center (DTDC) using Stochastic Reward Net (SRN).
Background:
Modern Data Center (DC)s are the core of the whole information ecosystem which provides advanced computing capabilities and modern service paradigms for enterprise businesses. New computing service paradigms rapidly coming in industry such as cloud computing, mobile computing, big data etc., urge a revolutionary development of IT infrastructures especially in DC. To provide uninterrupted online services and powerful computing
capabilities, DCSs are required to constantly operate over time. In this regard, the concepts of virtualization and software-defined environment are the core approach and promising solution. The virtualized and software-defined DCSs are capable of creating excessive
computing power and resources, high flexibility and agility in operation management, and
more importantly they are capable of confronting with a variety of security concerns and
system failures to avoid any interruptions for ordinary business processes and to assure high
availability and continuity of information resources flowing within an organization. Therefore, fault tolerance and disaster tolerance are critical demands for DCSs in practice in order
to achieve high availability and to avoid service disruptions. Assessing such DCSs with
fault tolerance and disaster tolerance under different availability metrics is of paramount
importance to help provide a prediction base of system availability to information system
administrators and providers in order to design and construct efficient and high available
data centers.
Key problems:
Existing approaches on availability assessment of DCSs attempt to model DCSs with
simplification and do not provide functionalities to capture and analyze correctly dynamic
and dependable behaviors within a system. Furthermore, although fault tolerance has been
taken into account in modeling and analysis of DCSs but there is lack of efforts to account
i
for disaster tolerance. In addition, it is critical to describe the DCSs as much as possible in
modeling, but the existing models are prone to capture complex interactions and dependencies within a system using conventional modeling approach.
Approach:
We develop analytical stochastic models with a comprehensive and high fidelity modeling approach. We attempt to take into account the stated problems in our models with
reasonable simplification. Through analyzing the proposed models under availability metrics, the solutions of the problems could be found.
Broad impact:
Research outcomes comprising of stochastic models and system availability assessment
of typical DCSs can be beneficial for variety of businesses in design and construction not
only of conventional data centers but also of contemporary systems including Infrastructure
as a Service (IaaS) in cloud computing systems, Software Defined Data Center (SDDC),
DTDC etc.

목차 (Table of Contents)

• ABSTRACT i
• ACKNOWLEDGMENTS iii
• LIST OF FIGURES x
• LIST OF TABLES xi
• ABBREVIATIONS xiii
• CHAPTER 1 INTRODUCTION 1
• 1.1 Motivation 1
• 1.2 Problem Statements 3
• 1.3 Objectives and Methodology 4
• 1.4 Contributions 7
• 1.5 Dissertation Organization 8
• CHAPTER 2 PRELIMINARIES 10
• 2.1 Chapter Outline 10
• 2.2 Background 11
• 2.2.1 Fundamental Concepts of Data Center Dependability 11
• 2.2.2 System Availability Definition 14
• 2.2.3 Taxonomy of Availability Modeling and Evaluation 17
• 2.2.4 Stochastic Reward Nets 20
• 2.2.5 Availability Analyses 22
• 2.3 Literature Review 24
• 2.3.1 Server Virtualization 24
• 2.3.2 Software Rejuvenation 25
• 2.3.3 Server Rejuvenation 26
• 2.3.4 Data Center’s High Availability, Fault and Disaster Tolerance 27
• 2.3.5 Data Center Network 32
• 2.3.6 Software Defined Network 34
• CHAPTER 3 A VIRTUALIZED SERVERS SYSTEM 37
• 3.1 Chapter Outline 37
• 3.2 Introduction 37
• 3.3 System Description of a VSS 39
• 3.3.1 System Architecture 39
• 3.3.2 Failure Modes and Recovery Behaviors 39
• 3.3.3 Assumptions 42
• 3.4 SRN Models of the VSS 43
• 3.4.1 System Model 43
• 3.4.2 Hosts and SAN Submodels 43
• 3.4.3 VMM Models with Time-based Rejuvenation 45
• 3.4.4 VM Models with Time-based Rejuvenation 47
• 3.5 Numerical Analysis Results of the VSS 50
• 3.5.1 Steady State Availability Analysis 51
• 3.5.2 Transaction Loss 52
• 3.5.3 Sensitivity Analysis 53
• 3.6 Discussion 56
• 3.7 Chapter Summary 57
• CHAPTER 4 A DATA CENTER NETWORK 58
• 4.1 Chapter Outline 58
• 4.2 Introduction 59
• 4.3 DCell-based Network Topologies of Typical DCNs 61
• 4.3.1 Network Topologies 61
• 4.3.2 System Behaviors and Assumptions 62
• 4.4 SRN Models of the DCNs 64
• 4.4.1 System Models 64
• 4.4.2 SRN Models of Hosts, VMs and Switches 65
• 4.4.3 SRN Models of a Standalone DCell0 67
• 4.4.4 System Model Integration 71
• 4.5 Numerical Analysis Results of the DCNs 72
• 4.5.1 Steady State Analysis of the DCNs 74
• 4.5.2 Sensitivity Analysis of the DCNs 77
• 4.6 Chapter Summary 82
• CHAPTER 5 A SOFTWARE DEFINED NETWORK 84
• 5.1 Chapter Outline 84
• 5.2 Introduction 85
• 5.3 A Software Defined Network 86
• 5.3.1 A Study-case of SDN Architecture 86
• 5.3.2 Incorporated failure modes of the SDN 87
• 5.3.3 Study Scenario 89
• 5.4 Hierarchical Models of the SDN 89
• 5.4.1 Hierarchical Architecture 89
• 5.4.2 Reliability Graphs 90
• 5.4.3 Host Model 93
• 5.4.4 Switch Model 94
• 5.4.5 SRN model of storage subsystem 97
• 5.5 Numerical Results and Discussion 98
• 5.5.1 Steady State Availability Analysis 98
• 5.5.2 Sensitivity Analysis 99
• 5.5.3 Discussion 105
• 5.6 Conclusions 106
• CHAPTER 6 A DISASTER TOLERANT DATA CENTER 107
• 6.1 Chapter Outline 107
• 6.2 Introduction 108
• 6.3 System Architecture of a DTDC 111
• 6.3.1 Architecture Description 111
• 6.3.2 Disaster Tolerant Configurations 112
• 6.3.3 Failure Modes and Recovery Behaviors of the DTDC 113
• 6.3.4 Assumptions 119
• 6.4 SRN Models of the DTDC 122
• 6.4.1 System Model 122
• 6.4.2 SRN Models of Disaster Occurrence 122
• 6.4.3 SRN Model of a Backup Server 125
• 6.4.4 SRN Models of Hosts and NASs 125
• 6.4.5 SRN Models of Standalone VM Subsystems in a Data Center 127
• 6.4.6 SRN Models of VM Subsystems with Fault-Tolerance in Data Centers 129
• 6.4.7 A SRN Model of VM Transmission in a DTDC with Disaster Tolerance and Fault Tolerance 132
• 6.5 Numerical Results 135
• 6.5.1 Steady-state Availability and Downtime Cost Analyses 139
• 6.5.2 Sensitivity Analysis 143
• 6.5.3 Discussion 147
• 6.6 Conclusion 149
• CHAPTER 7 CONCLUSION AND FUTURE WORK 150
• 7.1 Dissertation Conclusion 150
• 7.2 Future Work 153