Run-time Models for Self-managing Systems and Applications (eBook)

Preface 6
Contents 9
Stochastic Analysis and Optimization of Multiserver Systems 10
Introduction 10
Technical Preliminaries 12
Generic Model Description 12
Mathematical Definitions and Results 13
Boundary Value Problems 16
Stability and Throughput 17
Stochastic Decomposition 18
Stochastic Process Limits 23
Decentralized Control and Dynamics 26
Conclusions 28
References 29
On the Selection of Models for Runtime Prediction of System Resources 34
Introduction 34
Background 36
Prediction in Noisy Contexts 37
Statistical Properties of Time Series 38
Alternatives Choices for a Prediction Model 41
Alternatives in a Prediction Mechanism 41
Monitored and Filtered Time Series 42
Static and Dynamic Parameters 43
Results 45
Sensitivity to the Noise Component 45
Sensitivity Analysis 48
Conclusions 51
References 51
Estimating Model Parameters of Adaptive Software Systems in Real-Time 54
Introduction 54
Related Work 56
Main Results and Contribution 57
System Description 57
Static Performance Model Estimation 58
Extended Kalman Filter Design 59
Experimental Setup 62
Simulated Adaptive Software 62
Estimating Expected Values of Parameters 63
Traffic Generator 63
System Architecture and Hardware 64
Problems with Multiple Classes of Workload 64
Low Variation in Workload (LVW) Experiment 65
Step-Change in System Parameters (SSP) Experiment 67
Problem Analysis 70
Modified Filter Design & Improved Results
Insights for Choosing N and Each li 76
Improved LVW Experiment Results 77
Improved SSP Experiment Results 77
Conclusion 79
References 79
A Control-Theoretic Approach for the Combined Management of Quality-of-Service and Energy in Service Centers 81
Introduction 81
Problem Statement and Notation 84
LPV State Space Models: Identification and Validation 85
LPV Model Identification Approach 85
Testbed Setting and Experimental Results 87
Dynamic Analysis of the QoS/Energy Trade-off 89
Model Predictive Control for Performance Evaluation 90
Analysis for Optimal Performance 92
Cost Function Definitions 92
Dynamic Equality Constraints 93
Input and Performance Inequality Constraints 94
LPV-MPC Optimisation Problem 94
Simulation Results 95
QoS/Energy Trade-off Analysis 97
Time Domain Analysis 98
Concluding Remarks 101
References 102
The Emergence of Load Balancing in Distributed Systems: the SelfLet Approach 105
Introduction 106
The SelfLet Approach 107
Offered Services 107
Autonomic Policies 109
SelfLet Internal Architecture 109
Performance Model of a SelfLet System 111
Network Topology 111
SelfLet Workload 111
Behaviour Performance Model 113
Behaviour Utilization 114
State of a SelfLet 115
Autonomic Policies for Load Balancing 115
Prediction Model 116
Utilization Thresholds 116
Actions and Their Costs 117
Change Service Implementation 117
Service Redirect 118
Service Teach 119
General Policy 120
Experimental Results 124
Related Work 126
Conclusion 129
References 130
Run Time Models in Adaptive Service Infrastructure 133
Introduction 133
Setting the Context 135
The Process View: the Role of Models at Run Time 137
Instantiating the Process Model: 3 Scenarios 139
The PLASTIC Project 139
The CONNECT Project 152
The PFM Project 155
Concluding Remarks 157
References 158
On the Modeling and Management of Cloud Data Analytics 161
Introduction 161
Data Analysis at a Large Scale 163
Parallel DBMS 163
Map-Reduce Framework 164
Hadoop 165
Hive and Pig 166
Dryad 166
Performance Management 167
Fine-Grained and Embarrassingly Parallel Structure of Data Analytic Computations 167
Opportunity for Better Scheduling 167
Job Profiling 168
Large Scale, Commodity-Hardware Infrastructure 170
Scheduling and Workload Performance Modeling Must Recognize that Frequent Failure Is the Norm 170
Dataflow Resilience Issues 171
Scheduling and Resource Allocation Must Increase Data Locality of Tasks to Prevent Network Overload 171
Leveraging Low Latency Storage Technology 173
Management Objective to Provide User-Level Performance Guarantees at Low Cost 173
Defining User-Level SLA and SLA Management Issues 174
Fine Grained Resource Sharing 174
Modeling and Optimization 175
Inventory Modeling 176
Modeling Data Analytics 177
Modeling Map-Reduce Workflows 178
Conclusion 179
References 179