Performance Benchmark And Time-Constraint Violation Rate Of Embedded Systems

With the spread of the application domain, embedded system products have become the main stream of the digital products in the information era. Each product is unique and closely related with its specific application. The present performance evaluation tools can not satisfy the increasing application demands. In this thesis, a performance benchmark is introduced in detail.The performance of embedded system in this thesis means a system's capability to work. The three methods to evaluate the performance are analysis, simulation and measurement. Belongs to measurement method, the performance benchmark is a well defined task or a set of well defined tasks, which is a test used to evaluate a computer system or component. As it is flexible, low cost and be convenient to install and use, performance benchmark is suitable to measuring the working abilities of embedded systems.A new workload classification method and a new benchmark, Tjbench, is introduced by comparing and analysis various kinds of present performance benchmarks and by concluding their advantages and disadvantages, such as SPEC, Rhealstone, Lmbench, Mibench. This classification method is suitable for embedded application domain. This benchmark suite can define the workload according to the demand of embedded system and measure the working performance of different system.Having confirmed the workload, the idea of time constraint violation rate is introduced to compare and analyze the real-time performance of a system, which is an important ingredient influencing the overall performance of a system. A simplified task relation module is constructed to get the time constraint violation rate by system calls and data statistic. In this thesis, three different OSes, Linux, Red Hat Linux and RT-Linux are chosen. These experimental data illustrate their real-time abilities are different a lot from each other. All this shows to evaluate real-time systems by time constraint rates is feasible.