Model-Based Design for Embedded Systems

Gabriela Nicolescu, Pieter J. Mosterman

• Publisher: CRC Press
• Year: 2018
• Print ISBN: 9781138114722
• eBook ISBN: 9781351834711
• Edition: 1
• Topic: Computer Science
• Subtopic: Computer Engineering
• Index: Computer Science

The demands of increasingly complex embedded systems and associated performance computations have resulted in the development of heterogeneous computing architectures that often integrate several types of processors, analog and digital electronic components, and mechanical and optical components—all on a single chip. As a result, now the most prominent challenge for the design automation community is to efficiently plan for such heterogeneity and to fully exploit its capabilities.A compilation of work from internationally renowned authors, Model-Based Design for Embedded Systems elaborates on related practices and addresses the main facets of heterogeneous model-based design for embedded systems, including the current state of the art, important challenges, and the latest trends. Focusing on computational models as the core design artifact, this book presents the cutting-edge results that have helped establish model-based design and continue to expand its parameters.The book is organized into three sections: Real-Time and Performance Analysis in Heterogeneous Embedded Systems, Design Tools and Methodology for Multiprocessor System-on-Chip, and Design Tools and Methodology for Multidomain Embedded Systems. The respective contributors share their considerable expertise on the automation of design refinement and how to relate properties throughout this refinement while enabling analytic and synthetic qualities. They focus on multi-core methodological issues, real-time analysis, and modeling and validation, taking into account how optical, electronic, and mechanical components often interface. Model-based design is emerging as a solution to bridge the gap between the availability of computational capabilities and our inability to make full use of them yet. This approach enables teams to start the design process using a high-level model that is gradually refined through abstraction levels to ultimately yield a prototype. When executed well, model-based design encourages enhanced performance and quicker time to market for a product. Illustrating a broad and diverse spectrum of applications such as in the automotive aerospace, health care, consumer electronics, this volume provides designers with practical, readily adaptable modeling solutions for their own practice.

Part I

Real-Time and Performance
Analysis in Heterogeneous
Embedded Systems

1

Performance Prediction of Distributed Platforms

Lothar Thiele and Simon Perathoner

CONTENTS

1.1 System-Level Performance Analysis

1.1.1 Distributed Embedded Platforms

1.1.2 Role of Performance Analysis in the Design Process

1.1.3 Approaches to Performance Analysis

1.2 Application Scenario

1.3 Representation in the Time Domain

1.3.1 Arrival and Service Functions

1.3.2 Simple and Greedy Components

1.3.3 Composition

1.4 Modular Performance Analysis with Real-Time Calculus

1.4.1 Variability Characterization

1.4.2 Component Model

1.4.3 Component Examples

1.4.4 System Performance Model

1.4.5 Performance Analysis

1.4.6 Compact Representation of VCCs

1.5 RTC Toolbox

1.6 Extensions

1.7 Concluding Remarks

Acknowledgments

References

1.1 System-Level Performance Analysis

One of the major challenges in the design process of distributed embedded systems is to accurately predict performance characteristics of the final system implementation in early design stages. This analysis is generally referred to as the system-level performance analysis. In this section, we introduce the relevant properties of distributed embedded systems, we describe the role of the system-level performance analysis in the design process of such platforms, and we review different analysis approaches.

1.1.1 Distributed Embedded Platforms

Embedded systems are special-purpose computer systems that are integrated into products such as cars, telecommunication devices, consumer electronics, and medical equipment. In contrast to general-purpose computer systems, embedded systems are designed to perform few dedicated functions that are typically known at the time of design. In general, the knowledge about the specific application domain and the behavior of the system is exploited to develop customized and optimized system designs. Embedded systems must be efficient in terms of power consumption, size, and cost. In addition, they usually have to be fully predictable and highly dependable, as a malfunction or a breakdown of the device they may control is in general not acceptable.

The embedding into large products and the constraints imposed by the environment often require distributed implementations of embedded systems. In addition, the components of a distributed platform are typically heterogeneous, as they perform different functionalities and are adapted to the particular local environment. Also the interconnection networks are often not homogeneous, but may be composed of several interconnected subnetworks, each one with its own topology and communication protocol. The individual processing nodes are typically not synchronized. They operate in parallel and communicate via message passing. They make autonomous decisions concerning resource sharing and scheduling of tasks. Therefore, it is particularly difficult to maintain a global-state information of the system.

Many embedded systems are reactive systems that are in a continuous interaction with their environment through sensors and actuators. Thus, they often have to execute at a pace determined by their environment, which means that they have to meet real-time constraints. For these kinds of systems, the predictability in terms of execution time is as important as the result of the processing itself: a correct result arriving later (or even earlier) than expected is wrong.

Based on the characteristics described above, it becomes apparent that heterogeneous and distributed embedded real-time systems are inherently difficult to design and to analyze, particularly, as not only the availability and the correctness of the processed results, but also the timeliness of the computations are of major concern.

1.1.2 Role of Performance Analysis in the Design Process

Reliable predictions of performance characteristics of a system such as end-to-end delays of events, memory demands, and resource usages are required to support important design decisions. In particular, the designer of a complex embedded system typically has to cope with a large design space that is given by the numerous alternatives for partitioning, allocation, and binding in the system design. Thus, he or she often needs to evaluate the performance of many design options in order to optimize the trade-offs between several design objectives. In such a design space exploration, the performance analysis plays a crucial role, as can be seen in Figure 1.1.

FIGURE 1.1
Performance analysis in the design space exploration cycle.

Methods and tools for expedient and reliable performance analyses of system specifications at a high abstraction level are not only needed to drive the design space exploration but also for verification purposes. In particular, they permit to guarantee the functionality of a system in terms of real-time constraints before much time and resources are invested for its actual implementation.

1.1.3 Approaches to Performance Analysis

The need for accurate performance predictions in early design stages has driven research for many years. Most of the approaches for performance analysis proposed so far can be broadly divided into two classes: simulation-based methods and analytic techniques. There are also stochastic methods for performance analysis; however, we will not discuss them further in this context.

Simulation-based methods for performance estimation are widely used in industry. There are several commercial tools that support cycle-accurate cosimulation of complete HW/SW systems. Besides commercial tool suites, there also exist free simulation frameworks that can be applied for performance estimation, such as SystemC [9].

The main advantage of simulation-based performance estimation approaches is their large and customizable modeling scope, which permits to take into account various complex interactions and correlations in a system. In addition, in many cases, the same simulation environment can be used for both function and performance verifications. However, most simulation-based performance estimation methods suffer from insufficient corner-case coverage. This means t...

Table of contents

• Cover
• Half Title
• Title Page
• Copyright Page
• Table of Contents
• Preface
• Introduction
• Contributors
• Part I Real-Time and Performance Analysis in Heterogeneous Embedded Systems
• Part II Design Tools and Methodology for Multiprocessor System-on-Chip
• Part III Design Tools and Methodology for Multidomain Embedded Systems
• Index