Real-Time Simulation Technologies: Principles, Methodologies, and Applications

Katalin Popovici, Pieter Mosterman

• Publisher: CRC Press
• Year: 2017
• Print ISBN: 9781138077553
• eBook ISBN: 9781351833394
• Topic: Technology & Engineering
• Subtopic: Electrical Engineering & Telecommunications
• Index: Technology & Engineering

Real-Time Simulation Technologies: Principles, Methodologies, and Applications is an edited compilation of work that explores fundamental concepts and basic techniques of real-time simulation for complex and diverse systems across a broad spectrum. Useful for both new entrants and experienced experts in the field, this book integrates coverage of detailed theory, acclaimed methodological approaches, entrenched technologies, and high-value applications of real-time simulation—all from the unique perspectives of renowned international contributors.

Because it offers an accurate and otherwise unattainable assessment of how a system will behave over a particular time frame, real-time simulation is increasingly critical to the optimization of dynamic processes and adaptive systems in a variety of enterprises. These range in scope from the maintenance of the national power grid, to space exploration, to the development of virtual reality programs and cyber-physical systems. This book outlines how, for these and other undertakings, engineers must assimilate real-time data with computational tools for rapid decision making under uncertainty.

Clarifying the central concepts behind real-time simulation tools and techniques, this one-of-a-kind resource:

• Discusses the state of the art, important challenges, and high-impact developments in simulation technologies
• Provides a basis for the study of real-time simulation as a fundamental and foundational technology
• Helps readers develop and refine principles that are applicable across a wide variety of application domains
  

As science moves toward more advanced technologies, unconventional design approaches, and unproven regions of the design space, simulation tools are increasingly critical to successful design and operation of technical systems in a growing number of application domains. This must-have resource presents detailed coverage of real-time simulation for system design, parallel and distributed simulations, industry tools, and a large set of applications.

Section IV

Tools and Applications

17 Toward Accurate Simulation of Large-Scale Systems via Time Dilation

James Edmondson and Douglas C. Schmidt

CONTENTS

17.1 Introduction

17.2 Background

17.2.1 Formal Composition Techniques

17.2.2 Simulation Techniques

17.3 Motivating Scenarios

17.3.1 Ethernet Capture Effect

17.3.2 Application Services

17.3.3 Large-Scale Systems

17.4 Applying Time Dilation with Diecast

17.4.1 Overview

17.4.1.1 Time Dilation Factor (TDF) and Scale Factor (SF)

17.4.1.2 Paravirtualized vs. Fully Virtualized VMs

17.4.1.3 CPU Scheduling

17.4.1.4 Network Emulation

17.4.2 Application to Motivating Scenarios

17.4.3 Future Work

17.5 Addressing Issues of Time Dilation in Physical Memory

17.5.1 Overview

17.5.2 Solutions

17.5.2.1 More Memory or More Hosts

17.5.2.2 Memory Emulation

17.6 Concluding Remarks

References

17.1 Introduction

Distributed systems, particularly heterogeneous systems, have been historically hard to validate [1]. Even at small scales—and despite significant efforts at planning, modeling, and integrating new and existing systems into a functional system-of-systems—end users often experience unforeseen (and often undesirable) emergent behavior on the target infrastructure. Some types of unexpected emergent behaviors include unwanted synchronization of distributed processes, deadlock and starvation, and race conditions in large-scale integrations or deployments [2].

Deadlock and starvation are not just limited to large-scale systems and can occur when connecting just a few computers or computer systems together. Phenomenon such as the Ethernet Capture Effect [2,3] (which is a type of race condition involving a shared bus and accumulating back-off timers on resending data) once occurred on networks as small as two computers, despite decades of previous protocol use and extensive modeling. If problems like this can occur during small integrations or technology upgrades, the challenges of integrating large-scale systems containing thousands of computers, processing elements, software services, and users are even more daunting. Ideally, all technological upgrades and new protocols could be tested on the actual target infrastructure at full scale and speeds, but developers and system integrators are often limited to testing on smaller-scale testbeds and hoping that the behavior observed in the testbeds translates accurately to the target system.

Consequently, what we need are technologies and methodologies that support representative “at-scale” experiments on target infrastructure or a faithful simulation, including processor time and disk simulation, as well as network simulation. These technologies and methodologies should allow application developers to incorporate their application or infrastructure software into the simulator unmodified, and so they run precisely as expected on the target system. There are many network simulators available for use—some of which we discuss in this chapter—but we also explore a new simulation technology that was introduced by Gupta et al. and is called time dilation [4].

The term time dilation has roots in the theory of relativity, pioneered by Albert Einstein in the early twentieth century [5]. In physics, time dilation is a set of phenomena that describe how two observers moving through space relative to each other or at different positions relative to objects with gravitational mass will observe each other as having erroneous clocks, even if the clocks are of identical construction. Relative velocity time dilation, the phenomenon described by two bodies observing each other’s clocks while moving at different velocities, is the best parallel for the definitions and usage of the terminology in the work on simulation by Gupta et al.

Gupta et al. specifically coin the term time dilation in simulation to describe the process of altering time sources, clocks, and disk and network transactions to allow accurate simulation of multiple virtual machines (VM) on a single host, and we use this new definition throughout this chapter. This new usage fits with the original definitions by Einstein since a simulator and the actual operating system running the simulator will see time flowing at different rates due to context switching and other modern operating system techniques, despite both using equivalent clock mechanisms. The time dilation mechanism in the simulation context attempts to correct this clock drift from the simulator and operating system perspective to allow for closer approximation of target behavior by simulated tests.

Simulations based on time dilation allow system integrators and planners to run unmodified executables, services, and processing elements to accurately emulate CPU, network, disk, and other resources for large-scale systems in much smaller testbeds. A prototype of this time dilation technology called DieCast [6] has been implemented by researchers at the University of California at San Diego. This chapter explores the benefits of this technology to date, summarizes what testers must consider when using DieCast, and describes future work necessary to mature time dilation techniques and tools for simulation of large-scale distributed systems. This chapter presents a survey of the work to date by Gupta et al. in the application of time dilation to simulation in DieCast. It also motivates future work on memory management considerations for time dilation and the need for additional improvements to conventional time dilation implementations to address particular types of race conditions (such as the Ethernet Capture Effect) that may not be covered by time dilation.

17.2 Background

In general, large-scale distributed systems are difficult to validate. Though some simulators, such as USSF described in Section 17.2, can emulate networks consisting of millions of nodes, simulation technologies that scale to this level often deviate from the performance and characteristics of target large-scale systems. This section describes related work on validating distributed systems and summarizes the pros and cons of current validation techniques with respect to their ability to address the role of time dilation in large-scale system validation. We divide related background material into two main areas: formal composition and simulation techniques.

17.2.1 Formal Composition Techniques

Formal composition is typically associated with modeling a target system in a computer-aided manner that ensures the distributed system is validated based on validated components and will execute properly on the target system [7]. When constructing mission and safety-critical distributed real time and embedded systems, such as flight avionics computers and nuclear power plant control systems, software developers often use formal composition techniques and tools, such as Step-Wise Refinement [8], Causal Semantics [9], Behavioral Modeling [10], and Object Modeling Technique [7], to validate their software before it goes into production.

Formal composition techniques are often time consuming to validate, however, and can be tightly coupled to a particular development context and domain. Moreover, many formal composition methods require developers to model everything (e.g., from processors to operating systems to the application logic)...

Table of contents

• Cover
• Half title
• Title Page
• Copyright Page
• Table of Contents
• Preface
• Editors
• Contributors
• Introduction
• SECTION I Basic Simulation Technologies and Fundamentals
• SECTION II Real-Time Simulation for System Design
• SECTION III Parallel and Distributed Real-Time Simulation
• SECTION IV Tools and Applications
• Index