Energy consumption is increasingly affecting battery life and cooling for real- time systems. Dynamic Voltage and frequency Scaling (DVS) has been shown to substantially reduce the energy consumption of uniprocessor real-time systems. It is worthwhile to extend the efficient DVS scheduling algorithms to distributed system with dependent tasks. The dissertation describes how to extend several effective uniprocessor DVS schedul- ing algorithms to distributed system with dependent task set. Task assignment and deadline assignment heuristics are proposed and compared with existing heuristics concerning energy-conserving performance. An admission test and a deadline com- putation algorithm are presented in the dissertation for dynamic task set to accept the arriving task in a DVS scheduled real-time system. Simulations show that an effective distributed DVS scheduling is capable of saving as much as 89% of energy that would be consumed without using DVS scheduling. It is also shown that task assignment and deadline assignment affect the energy- conserving performance of DVS scheduling algorithms. For some aggressive DVS scheduling algorithms, however, the effect of task assignment is negligible. The ad- mission test accept over 80% of tasks that can be accepted by a non-DVS scheduler to a DVS scheduled real-time system.

GlobalEye is an aircraft by Saab that provides surveillance of air, sea, and land. In order to train on the system present in this aircraft a Mission Training System, MTS, is used. The MTS consists of the Command and Control, C2, system that is present in the aircraft as well as a scenario simulator. This master thesis is focused on evaluating the performance of the distributed real-time system that is the MTS in terms CPU and memory usage of its Java processes. These Java processes are divided across four servers and two types of workstations.  Two sorts of tests were conducted; workload model testing and user scenario testing. Both tests were conducted multiple times with different amount of workload. The workload was defined as a number of air and sea targets added to the scenario simulator. The data was collected by running bash scripts on each machine that saved the CPU and memory output from the Linux top command. Java VisualVM was used to examine a selection of processes further. In Java VisualVM the heap size, garbage collector pauses, and CPU time of the methods were the main focus of interest. The result from the workload model testing showed that a large majority of the processes handled an increased workload in terms of memory usage well. When it comes to the CPU usage four processes with significantly increasing CPU values were examined further in Java VisualVM. These processes were MMI processes and processes involved in data transfer. No problems were noted regarding the garbage collection and heap sizes. The longest pause, across all processes, caused by garbage collection was 29.7ms. The methods with highest amount of CPU self time were also identified for the processes. One observation was that a method related to map rendering, for the MMI processes, showed decreases in CPU time as the workload increased. The result from the user scenario testing showed that the MMI of the scenario simulator suffers delays dependant on the number of targets present in the scenario. The Java process of the MMI did increase its CPU usage significantly as the targets increased. In Java VisualVM it was observed that the garbage collector for the scenario simulator MMI process was more extensive during these tests, with the longest pause being 104ms. Lastly when it comes to the CPU self time of the methods within the MMI process, one error related method was identified.

Dans cette thèse, nous analysons les réseaux des systèmes temps-réel distribués et plus particulièrement ceux des domaines de l’avionique et de l’automobile. Nous nous sommes focalisés sur deux protocoles : « Avionic Full DupleX Switched Ethernet » (AFDX), « Audio Vidéo Bridging Ethernet » (AVB). Dans ces domaines critiques, le déterminisme du réseau doit être garanti. Il consiste, notamment, en la détermination d’une borne garantie du délai de bout en bout de traversée du réseau pour chaque trame ; et un dimensionnement des files d’attente des trames suffisamment grand pour garantir qu’aucune d’entre elle ne débordera et ainsi, éviter toute perte de trame.Il existe plusieurs méthodes pour l’évaluation des délais et nous avons, principalement, travaillé sur la méthode « Forward end-to-end delay Analysis » (FA). FA avait déjà été définie avec la politique d’ordonnancement « First-In-First-Out » dans le contexte de l’AFDX. Nous sommes repartis de cette approche, nous l’avons reformulé et généralisé à n’importe quel réseau Ethernet commuté. Nous l’avons aussi étendu aux priorités statiques et au protocole AVB et sa politique de service « Credit Based Shaper ». Pour chaque contribution, des démonstrations formelles ont été présentées et une expérimentation incluant une comparaison de FA avec les principales approches d’évaluation sur un exemple industriel. Finalement, nous avons développé et démontré formellement une approche pour le dimensionnement des files d’attente en termes de nombre de trames. Cette approche a été expérimentée également sur une configuration industrielle. / In this thesis, we analyze networks in the context of distributed real-time systems, especially in the fields of avionics, with “Avionics Full DupleX Switched Ethernet” (AFDX), and automobile, with “Audio Video Bridging Ethernet” (AVB). For such applications, network determinism needs to be guaranteed. It involves, in particular, assessing a guaranteed bound on the end-to-end traversal time across the network fr each frame; and dimensioning the buffers in order to avoid any loss of frame because of a buffer overflow.There are several methods for worst-case delay analysis, and we have mainly worked on the “Forward end-to-end Delay Analysis” (FA) method. FA had already been developed for “First-In-First-Out” scheduling policy in the AFDX context, so we generalized it to any Switched Ethernet network. We have also extended it to handle static priorities and the AVB protocol, shaping policy named “Credit Based Shaper” (CBS). Each contribution has been formaly proved and experiments have been led on industrial configurations. For our experimentations, we have compared our results with the results of competing approaches. Finally, we have developed and formally demonstrated an approach for buffer dimensioning in terms of number of frames. This approach has also been tested on an industrial configuration and has produced tight bounds.