Research On Time Predictability Of Communication Modules In Servant/Exe-Flow Model

Time predictability requires systems to estimate the execution time of real-time tasks and to meet the timing requirements. As for the strict time constrains, time pre-dictability becomes a key performance indicator for real-time systems. Minicore is a new microkernel based on Servant/Exe-Flow Model, in which the execution model and the storage model are independent of each other. With the purpose of hard real-time operating system, Minicore should fully be realized with a high time predictability capability. The implementation of a time predictable system requires a detailed design phase, which means that Minicore code needs an off-line analysis to prove its time pre-dictability. Conducting such a static analysis is a challenging task, since timing require-ments propagate from the top down to all the parts of the system, such as processor and instruction set architecture, language and compiler support, runtime system, scheduling, communication, etc.The highly modularity design of microkernel made Minicore rely heavily on the inter-servant communication module, because a servant is the basic unit of the operating system. As a result, in order to analyse its time predictability, it is inevitable to devise a time predictability analysis for Communication Modules in Servant/Exe-Flow Model. It is the Communication Modules that support Exe-Flow to flow between servants, re-questing some specific service and fulfilling the tasks. The only primitive for sending message is named "sendmsg", with three modes namely continuous mode, split mode and asynchronous mode.Our work is to explore the time predictability capability for Minicore. To begin with, we implement Minicore’s communication modules, estimate its WCET (Worst-Case Execution Time) and analyze its time predictability with the definition of CIPr (Configuration-Induced Timing Predictability). Static WCET analysis and a path-based algorithm have been adopted in our work. Applying to the communication modules in Minicore, the method includes four phrases:extract code snippets for each commu-nication mode, analyse communication control flow, analyse processor features, and calculate WCET finally. Section3provides data to support its future mechanism of timing predictability, and foundation for future work of time predictability analysis for Minicore.Next, by preferentially scheduling mini ports in the same address space. Lazy exe-flow direction is proposed in Section4, which takes advantage of the fact that there is significantly less overhead involving in context switch between mini ports in the same address space than in exe-flow direction to a mini port in another address space. We change the scheduler with the responsibility to schedule only the ready mini ports in the current address space, rather than ones in the whole system. When it finds it empty in the ready queue, it invokes a procedure called Director, choosing a servant to replace the current one. Minicore performs well with lazy exe-flow direction, and even better when the system workload is at a high level.