Research On The Time-Order Predictable Real-Time Operating System Model

Cyber-Physical Systems (CPS) are integrations of computation with physical pro-cesses. Because of the adaptability, automation, reliability, and feasibility of CPS, they have become a research hotspot in both the academic and the industry. Compared with the traditional real-time systems, CPS have strict requirements for composability and reliability. However, because of the concurrent, heterogeneous, interactive, and dis-tributed characteristics of CPS, it is in need to improve the system’s predictability for the purpose of guaranteeing the composability and reliability. This dissertation targets to define the predictability for CPS formally, and propose a new real-time operating systems (RTOS) model that can improve the predictability of the run-time temporal semantics.We define the predictability for CPS formally at first. Based on the character-istics and requirements of CPS, we argue that the I/O behaviors and the Computing behaviors need to be Time-Order predictable. We systematically analyze the source of uncertainties and propose metrics to measure the Time-Order predictability. The formal definitions of predictability are the foundations to the predictable RTOS work of this dissertation.We propose a Time-Order predictable RTOS model, i.e., the Predictable Servant/ Executive-Flow Model (PSEFM), to improve the predictability of RTOS by preserving the design-time Time-Order semantics during run-time:1. PSEFM uses the Rendezvous-Servant (R-Servant) and Synchronous-Servant (S-Servant) to handle the nondeterminism and determinism of a task, respectively. By this way, it is possible to express the nondeterminism explicitly and process the nondeterminism predictably.2. Based on the Logical Execution Time (LET) model, wepropose the SubLET mod-el to set logical time constraints for S-Servants. We present methods of combining the LET model with the Discrete Event (DE) model to support the LET seman-tics of S-Servants during run-time. A set of mechanisms to sort the concurrency events are proposed to improve the Time-Order predictability when processing nondeterminism. Based on the Time-Order constrains of PSEFM, we present an efficient strategy to check the safety of an event.3. Executive-Flow (EF) is the abstraction of the physical computing unit. PSEFM regards EF as an exclusive resource with physical time. Based on the Time-Order constraints of S-Servant and R-Servant, we propose predictable EF leading meth-ods for the multicore system.We propose and implement a PSEFM-based RTOS framework, PsefmOS. The e-valuation results show that PsefmOS has fewer run-time costs than EMachine which is used to support LET programming model. Then, we use PSEFM to implement an Unmanned Aerial Vehicles benchmark PapBench to test the effectiveness of PSEFM-based systems and the advantages of the Time-Order predictable system. The evalua-tion results show that, compared with the thread-based PapaBench, the PSEFM-based Papabench has the following advantages:(1) The PSEFM-based system supports the Time-Order semantics of programming model and has higher Time-Order predictabil-ity;(2) The PSEFM-based achieves higher least upper bound to processor utilization during run-time; (3) As long as tasks in the system are schedulable, the response time of a task is not affected by the increase or decrease of the task execution time and the number of tasks. Therefore, the PSEFM-based system performs better composability.