Study On Design And Optimization Of Communication Schemes In Embedded Real-Time Systems

Recently, embedded real-time systems have been applied to more and more fields,including household, industrial control, automotive, trains, aviation, etc. On one hand,with the increasing electrification of control systems, mechanical and hydraulic systems are being replaced by electronic components. On the other hand, more and more functionality and performance requirements(such as security and comfort) are imposed on the design of systems. These trends require the real-time systems should provide much stronger computing capabilities, which leads to a development towards more cores on a single processor and more processors in the system. This brings new requirement to the performance of communication schemes. The thesis explores and studies the communication schemes of real-time systems, then designs and optimizes the communication mechanisms among tasks on multicore platforms and the architecture of real-time networks. The main contributions are summarized as follows.(1) Data communication mechanisms for real-time applications in multicore platformsIn real-time systems, variables shared by tasks on different cores need to be protected by mechanisms that guarantee access in a mutual exclusive way. Multicore platforms are increasingly used in real-time embedded applications, due to their benefit in high-performance, low power consumption, etc. Because of the distribution property of multicore architecture, the design of communication schemes becomes more complicated.Currently, the communication schemes for multicore platforms can be classified into lockbased scheme and wait-free scheme. However, either of the schemes only has one advantage in low memory consumption or in system schedulability. The thesis proposes a novel scheme named FMCS(Flexible Multiprocessor Communication Scheme). This scheme assigns each shared variable the most suitable communication policy and collision handling policy through a heuristic algorithm proposed by us. FMCS provides for each task the formula of its worst-case response time(WCRT) which enables deterministic timing analysis for real-time systems. The experimental results show that FMCS dramatically reduces memory consumption as well as guaranteeing the system schedulability. The system can still be scheduled with FMCS even if the CPU utilization is higher than 90%.Therefore, FMCS is of high scalability.(2) The deployment of messages onto real-time networks with minimum additional delayWhen real-time systems are used as control systems, the latency of control path is vital to control precision. So the latency of control path should be limited in a given range. In model-based development, system properties can be verified in advance on function models by simulation or model checking. To ensure that the properties still hold for the final deployed system, the implementation into software tasks and communication messages should preserve the semantics properties of the model. In some cases, such a schedule is not feasible and the model should be modified by adding communication delays. The latency of message occupies a huge proportion in the latency of model. However, the study on message deployment in model-based deployment is lacking. The thesis builds a novel model named LLMD(Low-Latency Message Deployment). This model is a mathematical one based on the schedule of Flex Ray, a time-triggered communication scheme.LLMD can find the best message deployment solution, aiming at introducing least latency into critical control path, guaranteeing the deadlines of other paths and the system schedulability. The experimental results show that LLMD finds a message deployment solution which introduces as little additional latency as possible while guaranteeing system schedulability. The runtime of LLMD is relatively short.(3) The study of security scheme in real-time networksWith the increase of networked devices and electrification in avionics and automative systems, the accessibility of systems arises dramatically. This brings a series of security problems to real-time networks. The thesis proposes a scheme named SSS(Smart Security Scheme), which effectively protects the communication system from the threat of modification and fabrication and thus increases the security of real-time systems. SSS allows software implementation of security-related operations(including signing, verification and key generation) instead of hardware implementation, on light-loaded ECUs,and thus decreases hardware overhead. Meantime, we propose a design optimization model of SSS, which requires as few hardware units as possible by carefully scheduling the transmission of keys and messages and the execution of tasks while guaranteeing security and system schedulability. The model provides multiple security levels and supports designers select the most suitable security level according to the load of Flex Ray. The experimental results show that SSS finds the optimized solution within limited time and largely decreases hardware overhead, compared to existing Flex Ray security schemes.(4) The study of scalability in real-time networksAlong with the continuous increase of functionalities and performance of embedded systems, there is a rapid rise of the task number and communication data among tasks in real-time systems. This imposes higher requirements on the scalability of realtime networks. The thesis proposes HSRN, a highly scalable real-time network architecture. HSRN greatly improves the scalability of real-time networks through separating the system into multiple branches and broadcasting domains with switches. The switches in HSRN employ a proposed message-buffered structure. The structure allows the inter-branch messages to transmit at different slot positions of senders and receivers.Even if multiple messages from different branches are being sent to the same destination branch, the collision can be avoided by buffering messages in the switch. So the proposed switch architecture relaxes the constraint for message scheduling and thus improves system schedulability. In order to improve the load balance of different branches in HSRN,the thesis proposes a load-balancing algorithm. This algorithm can find an optimal solution of network splitting in a limited time. Thus, it improves the scalability of real-time network while guaranteeing the load balancing of different branches in HSRN. The experimental results show that HSRN can make a good tradeoff among the total number of the used slot, load balancing and the minimum switch buffer requirement.In summary, the thesis investigates several problems in the communication schemes of real-time systems, designs and optimizes the communication schemes for different purposes, and proposes several models and optimized design. Our works have their academic and practical value on promoting the advancement of research in communication schemes of real-time systems.