Project topics and materials
Get Complete Project Material File(s) Now! »
Network Experimentation Tools
A network experimentation tool is a framework that allows the experimenter to realize a given experiment. Several peripheral services can be added to this definition such as pre-definition service that help the experimenter to define the experiment or archiving/analyzing tools that simplify the usage of the output. Regarding the literature reviews, classification works consider two fundamental parameters: the implementation granularity and the realism level. Accordingly, we highlight five principal approaches to network experimentation: real world field trial, real world testbed, emulation testbed , simulation testbed and hybrid testbed.
Real world Field Trial
Real world field experiment consists of deploying an experimental subject through real world infrastructure in order to study its behavior, efficiency or limits. In general, experimenters use real computers and network equipment, where they rapidly deploy their subjects of study (protocols, hardware, software). This approach was one of the first validation methods in network research and industry because its initialization requirements are limited and it provides realistic results. Nevertheless, given the number and complexity of the factors involved in the current networks, real world fields display three limitations:
• Precision: the real world experiment is based on a black-box approach where the system is evaluated as a whole. Thus, it is difficult to isolate a part of the system in order to analyze its impact, behavior or performance. Consequently, debugging and profiling activities are incompatible with such method.
• Reproducibility: a real world experiment is by definition, achieved in an open context where real traffic and real equipment cohabit with experimental one. As a consequence, it is hard to control environmental parameters that may influence the experiment. Thus, the reproducibility of such experiments is limited and remains an open issue.
Network Experimentation Aspects
In previous sections, we classify five types of network experimentation tools, based on their realism level. Realism level and scalability can be used to classify experimentation tools. Table 2.1 summarizes a realism-based classification approach that focus on most common examples of each category. Nevertheless, the realism level remains one aspect of the experiment. A synthesis of the literature reviews highlights five major axes that define a given experimentation tool.
1. The architecture: the tool architecture summarizes the concepts that define the achievement of the tool. It includes the experiment base which defines how the experiment is described in the tool (components definition, topology, traffic description). The second aspect of the architecture is the component base. In fact, each component can be defined using realistic/simplified definition. Finally, the simplification aspect considers the user point of view. It is relative to the simplicity of the usage of the tool and not to the quality of the experiment.
2. Scientific experiment features: in order to be considered as a scientific experiment, the experimentation tool must address three features: the reproducibility, the revisability and the traffic quality. The reproducibility is the ability to reproduce the same experiment, expects to produce the same output and observes the same behavior. The revisability concerns the capacity of the experimenter to explore the studied system during the experimentation in order to extract meaningful results. The granularity of the exploration is important in the revisability of a given experiment (including dynamic debug and profiling). The traffic quality is an experimentation feature that may define the scientific quality. Thus, the ability of a given testbed to play real traces or to generate realistic traffic is an important feature.
Table of contents :
1 Introduction
1.1 Motivation and Objectives
1.2 Contributions Storyline
1.3 Thesis Structure
1.4 Publications
1.5 Project Deliverables
I Background
2 Network Experimentation
2.1 Introduction
2.2 Network Experimentation Tools
2.2.1 Real world Field Trial
2.2.2 Real world Testbed
2.2.3 Emulation Testbed
2.2.4 Simulation Testbed
2.2.5 Hybrid Testbed
2.3 Network Experimentation Aspects
3 Discrete Event Simulation
3.1 Introduction
3.2 Discrete-Event Simulation
3.2.1 Terminology and Components
3.2.2 The Principle
3.3 Parallel Discrete Event Simulation
3.3.1 Discrete Event Simulation Limits
3.3.2 Principles of Parallel Discrete Event Simulation
3.3.3 Parallel Simulation Model and Algorithms
4 Hardware Trends
4.1 Introduction
4.2 Evolution of Computing Chips
4.2.1 CPU: Historical Evolution and Trends
4.2.2 GPU: Historical Evolution and Trends
4.2.3 Emerging Solutions
4.2.4 Multi-Core accelerator
4.3 Parallel Programming: Models and API
4.3.1 Pthreads
4.3.2 OpenMP
4.3.3 MPI
4.3.4 CUDA
4.3.5 OpenCL
5 Related Work
5.1 Introduction
5.2 Large Scale Simulation
5.3 PDES Issues
5.3.1 Data Representation
5.3.2 Event Scheduling
5.4 Considerations Heterogeneous Computing Considerations
II Contributions
6 Cunetsim: An Experimentation Framework to Discover Scalability Horizons
6.1 Introduction
6.2 Fundamental Concepts
6.2.1 The Worker Pool
6.2.2 Separation Between an Event and Its Description
6.2.3 Massive Parallel Event Generation
6.3 Cunetsim Software Architectures
6.3.1 The Worker Design
6.3.2 The Master Design
6.3.3 Legacy Architecture for Multi-Core CPU
6.4 Comparative Performances Results
6.4.1 Simulation Runtime
6.5 Technical Challenges of GPU-based Simulation
6.5.1 Synchronization Challenge
6.5.2 Memory Management Challenge
6.5.3 Precision Issue
6.6 Configuration Issues of GPU-Oriented Simulation
6.6.1 Space Representation and partitioning
6.6.2 Tuning Parameters: Block Size as a Study Case
6.7 Conclusion
7 Hybrid Events Scheduler
7.1 Introduction
7.2 The Hybrid Scheduler
7.2.1 Model and Components
7.2.2 Scheduling Algorithms
7.3 Performance Evaluation
7.3.1 Scenario & Setup
7.3.2 Comparative Evaluation
7.3.3 Performance Analysis
7.4 Related Work
7.5 Discussion
7.6 Conclusion
8 General Purpose Coordinator-Master-Worker Model
8.1 Introduction
8.2 The General Purpose Coordinator-Master-Worker Model
8.2.1 Events Management: Description, Scheduling and Execution .
8.2.2 The Synchronization Mechanism of the GP-CMW Model
8.2.3 GP-CMW Communication Model
8.3 Comparative Evaluation
8.3.1 Comparative Performance Evaluation
8.3.2 Inherent Performance Evaluation
8.4 Related Work
8.5 Conclusion
9 Study Case of PADS Methodology Deployment: NS-3
9.1 Introduction
9.2 Overview
9.3 Events scheduling on NS-3
9.4 NS-3 Events scheduler extensions
9.4.1 The Explicit CPU Parallelism
9.4.2 The Implicit CPU Parallelism
9.4.3 The GPU Offloading
9.4.4 The Co-scheduler Approach
9.5 Comparative evaluation
9.5.1 Medium Load
9.5.2 High Load
9.6 Conclusion
III Conclusion
10 Conclusion
A Experimentation Methodology
A.1 Introduction
A.2 Scientific Experimentation
A.3 OpenAirInterface Experimentation methodology
A.3.1 OpenAirInterface Formal Experimentation Methodology
A.3.2 Methodology Implementation
A.4 Conclusion
B Résumé Étendue
Bibliography
