Master Thesis - Past Projects - Abstract

Evaluation of schedulers using the Enea Advanced Scheduling Framework and improvements of the framework software

ID: LiTH-IDA-Ex-04/024-SE

Every day new electronics are being introduced into everything we use. Systems become more and more complex. One category of systems that is growing rapidly, which builds the foundation of this thesis, are systems used for safety critical applications. These systems are often built up of many different processes that share common hardware resources. Examples of industries that use real-time theory when constructing embedded systems are car- and phone industries. It is also widely used in airplanes and medical applications. In many of these types of systems it is not only the result of a given computation that is important, but also when the result becomes available. Real-time theory is used to share system resources among different processes, as well as to help guarantee that the results from the processes are delivered within their temporal demands. In order for processes to be able to share resources there has to be some sort of mechanism that decides which process should get to use a specific hardware part during what time period, for example get to use the CPU. Such a mechanism is called a scheduler and one goal of this report is to evaluate two such schedulers to see if it is possible to create a rule of thumb when one scheduler is better than the other. Enea Epact has during the last two years developed a tool, the Enea Advanced Scheduling Framework (Enea ASF), to aid analysis of the temporal properties of a system. The tool can collect time measurements from the target platform automatically to verify and improve the results of the analysis. This thesis presents the enhancements to this framework made during the work on evaluating the schedulers, which includes a new graphical interface and the rewriting of the internal structure. By using tools like the Enea ASF it is possible to get early indication on how a system behaves and what type of hardware that is required to run the system. The ability to get early indications on what type of hardware that is required in a system, can greatly reduce a project costs. Such a tool can also pinpoint the areas that are most vital to optimize in order to minimize hardware costs. The result from the evaluation of the schedulers is that the EDF scheduler is much better at preventing deadline misses than the FPS scheduler, but uses more system resources in doing so. The difference in use of system resources is not very big though, the EDF scheduler utilizes about 0,008%/process more of the available time between each system tick than the FPS scheduler. This is partially compensated by the fact that the EDF scheduler causes less process preemptions.

Keywords: Real-time, Scheduling, RMA, Enea ASF, Embedded system

Author(s): Markus Malmgren

Contact: Jörgen Hansson