Edge Intelligence: From Theory to Practice

✍ Scribed by Javid Taheri, Schahram Dustdar, Albert Zomaya, Shuiguang Deng

• Publisher: Springer
• Year: 2023
• Tongue: English
• Leaves: 254
• Category: Library

✦ Synopsis

This graduate-level textbook is ideally suited for lecturing the most relevant topics of Edge Computing and its ties to Artificial Intelligence (AI) and Machine Learning (ML) approaches. It starts from basics and gradually advances, step-by-step, to ways AI/ML concepts can help or benefit from Edge Computing platforms.

The book is structured into seven chapters; each comes with its own dedicated set of teaching materials (practical skills, demonstration videos, questions, lab assignments, etc.). Chapter 1 opens the book and comprehensively introduces the concept of distributed computing continuum systems that led to the creation of Edge Computing. Chapter 2 motivates the use of container technologies and how they are used to implement programmable edge computing platforms. Chapter 3 introduces ways to employ AI/ML approaches to optimize service lifecycles at the edge. Chapter 4 goes deeper in the use of AI/ML and introduces ways to optimize spreading computational tasks along edge computing platforms. Chapter 5 introduces AI/ML pipelines to efficiently process generated data on the edge. Chapter 6 introduces ways to implement AI/ML systems on the edge and ways to deal with their training and inferencing procedures considering the limited resources available at the edge-nodes. Chapter 7 motivates the creation of a new orchestrator independent object model to descriptive objects (nodes, applications, etc.) and requirements (SLAs) for underlying edge platforms.

To provide hands-on experience to students and step-by-step improve their technical capabilities, seven sets of Tutorials-and-Labs (TaLs) are also designed. Codes and Instructions for each TaL is provided on the book website, and accompanied by videos to facilitate their learning process.

✦ Table of Contents

Preface
Acknowledgments
Contents
1 Distributed Computing Continuum Systems
1.1 Introduction
1.2 Related Work
1.3 System Management in the Cartesian Space
1.3.1 Cartesian Blanket
1.3.2 Computing Continuum Characteristics
1.3.3 Current Issues and Challenges
1.3.3.1 Reactive Management System
1.3.3.2 System's Stability Is Linked to the Infrastructure
1.3.3.3 Unknown System Derivatives
1.3.3.4 Lack of Causality Relations
1.4 System Management in the Markovian Space
1.4.1 Vision
1.4.1.1 System State
1.4.1.2 Markov Blanket
1.4.1.3 Equilibrium
1.4.1.4 Adaptation
1.4.1.5 An Illustrative Example
1.4.2 Learning
1.4.2.1 Design Phase Learning
1.4.2.2 Runtime Phase Learning
1.5 Use Case Interpretation
1.5.1 Application Description
1.5.2 SLOs as Application Requirements
1.5.3 Developing the DAG
1.6 Conclusion
References
2 Containerized Edge Computing Platforms
2.1 Containers vs. Virtual Machines
2.2 Container Engines
2.2.1 Docker
2.2.2 Podman
2.2.3 LXD
2.3 Container Orchestration Platforms
2.3.1 Self-Hosted vs. Managed Container Orchestration
2.3.2 Well-Known Container Orchestration Platforms
2.4 Kubernetes
2.4.1 Kubernetes Cluster
2.4.1.1 Control Plane Components
2.4.1.2 Node Components
2.4.2 Kubernetes Objects and Resource Types
2.4.3 Container Interfaces
2.4.3.1 Container Runtime Interface (CRI)
2.4.3.2 Container Network Interface (CNI)
2.4.3.3 Container Storage Interface (CSI)
2.4.4 Accessing and Managing Kubernetes Resources
2.4.4.1 Kubernetes Kubectl
2.4.4.2 Kubernetes Dashboard
2.5 Kubernetes SDKs
2.5.1 Kubernetes Python Client
2.5.1.1 Installing the Python Client
2.5.1.2 Using the Python Client
2.5.2 Kubernetes Java Client
2.5.2.1 Installing the Java Client
2.5.2.2 Using the Java Client
2.6 Summary
References
3 AI/ML for Service Life Cycle at Edge
3.1 Introduction
3.1.1 State of the Art
3.1.1.1 Wireless Networking
3.1.1.2 Service Placement and Caching
3.1.1.3 Computation Offloading
3.1.2 Grand Challenges
3.2 Al/ML for Service Deployment
3.2.1 Motivation Scenarios
3.2.1.1 The Heterogeneous Network
3.2.1.2 Response Time of Micro-Services
3.2.1.3 A Working Example
3.2.2 System Model
3.2.2.1 Describing the Correlated Micro-Services
3.2.2.2 Calculating the Response Time
3.2.3 Problem Formulation
3.2.4 Algorithm Design
3.2.4.1 Variables
3.2.4.2 The SAA-RP Framework
3.2.4.3 The GASS Algorithm
3.3 AI/ML for Running Services
3.3.1 System Description and Model
3.3.2 Algorithm Design
3.3.3 RL-Based Approach
3.4 AI/ML for Service Operation and Management
3.4.1 System Model
3.4.2 Problem Analysis
3.4.3 Dispatching with Routing Search
3.4.4 Scheduling with Online Policy
3.5 Summary
References
4 AI/ML for Computation Offloading
4.1 Introduction
4.2 AI/ML Optimizes Task Offloading in the Binary Mode
4.2.1 System Model
4.2.1.1 Local Execution Latency Evaluation
4.2.1.2 Task Offloading Latency
4.2.1.3 Battery Energy Consumption
4.2.1.4 Problem Formulation
4.2.2 Cross-Edge Computation Offloading Framework
4.3 AI/ML Optimizes Task Offloading the Partial Mode
4.3.1 System Model and Overheads
4.3.1.1 System Model
4.3.1.2 Overheads
4.3.2 Problem Formulation
4.3.3 Solution
4.3.3.1 Allocation of CPU Frequency and Power
4.3.3.2 Solution of Offloading Policy
4.3.3.3 Algorithm Analysis
4.4 AI/ML Optimizes Complex Jobs
4.4.1 System Model and Problem Formulation
4.4.1.1 A Working Example
4.4.1.2 Problem Formulation
4.4.2 Algorithm Design
4.4.2.1 Finding Optimal Substructure
4.4.2.2 Optimal Data Splitting
4.4.2.3 Dynamic Programming-Based Embedding
4.5 Summary
References
5 AI/ML Data Pipelines for Edge-Cloud Architectures
5.1 Introduction
5.2 State-of-the-Art Stream Processing Solutions for Edge-Cloud Architectures
5.3 Data Pipeline in Existing Platforms
5.4 Critical Challenges for Data Pipeline Solutions
5.5 MapReduce
5.5.1 Limitations of MapReduce
5.5.2 Beyond MapReduce
5.6 NoSQL Data Storage Systems
5.6.1 Apache Cassandra
5.6.2 Apache Flink
5.6.2.1 Flink Connectors
5.6.2.2 Flink Architecture
5.6.2.3 Flink Deployment Plan
5.6.3 Apache Storm
5.6.3.1 Storm Concepts
5.6.3.2 Storm Deployment Architecture
5.6.4 Apache Spark
5.6.4.1 Spark Architecture
5.6.4.2 Spark Execution Engine
5.7 Conclusion
References
6 AI/ML on Edge
6.1 Introduction
6.2 System Overflow
6.2.1 Caching on the Edge
6.2.2 Training on the Edge
6.2.3 Inference on the Edge
6.2.4 Offloading on the Edge
6.3 Edge Training
6.3.1 Architecture
6.3.2 Training Optimization
6.3.3 Federated Learning
6.4 Edge Inference
6.4.1 Model Design
6.4.2 Model Compression
6.4.2.1 Network Pruning
6.4.2.2 Quantization
6.4.2.3 Knowledge Distillation
6.5 Summary
References
7 AI/ML for Service-Level Objectives
7.1 SLO Script: A Language to Implement Complex Elasticity-Driven SLOs
7.1.1 SLOs and Elasticity
7.1.2 Motivation
7.1.3 Research Challenges
7.1.4 Language Requirements Overview
7.1.5 SLO Script Language Design and Main Abstractions
7.1.5.1 SLO Script Overview and Language Meta-Model
7.1.5.2 StronglyTypedSLO
7.1.5.3 Strongly Typed Metrics API
7.1.5.4 SLOC Object Model
7.2 A Middleware for SLO Script
7.2.1 Research Challenges
7.2.2 Framework Overview
7.2.2.1 Architecture
7.2.2.2 SLOC CLI
7.2.3 Mechanisms
7.2.3.1 Orchestrator-Independent SLO Controller
7.2.3.2 Provider-Independent SLO Metrics Collection and Processing Mechanism
7.2.4 Implementation
7.2.4.1 Orchestrator-Independent SLO Controller
7.2.4.2 Provider-Independent SLO Metrics Collection and Processing Mechanism
7.3 Evaluation
7.3.1 Demo Application Setup
7.3.2 Qualitative Evaluation
7.3.3 Performance Evaluation
7.4 Summary
References