片间综合化互连时间触发通信调度方法

未来先进的分布式综合化航空电子系统需要实现跨越计算体系结构的信息综合，对微小型智能器件间的综合化互连提出要求。通过构建一种具有开放式接口的芯片间综合化互连结构，提出一种相应的时间触发（TT）通信调度方法。首先建立片间综合化互连模型和时间触发流量传输模型，给出负载均衡的选径方法；随后依据传输路径中各级节点各流量的发送时间偏移量计算流量传输的等待时间可行值，利用遗传算法优化调整各芯片发送端口的调度表相位，更新各流量在发送端口的时间偏移量，缩短最坏情况下的最大等待时间，得到具有全局优化意义的时间触发调度表。与使用Yices等SMT形式化求解器的TT调度表生成方法相比，本方法不会出现长时间不停机而无法判定的问题，且不论对于对称或非对称结构，案例研究表明本方法的可调度规模至少增加30%。此外，以流量传输等待时间占周期的比例作为归一化的传输延迟度量，与既有的基于特征任务的调度方法相比，案例研究表明本方法得到的传输延迟最多仅为后者的2%。

Time-triggered communication scheduling method for off-chip integrated interconnection

The advanced distributed integrated avionics system needs to achieve information integration across computing architectures. Integrated interconnection among micro-smart devices is thus required. By constructing an off-chip integrated interconnection structure with the open interface, a corresponding time-triggered message scheduling method is proposed. First, the off-chip interconnection model and the Time-Triggered (TT) communication model are established, and the load balancing path selection method is given. Then, the feasible value of the waiting time is calculated according to the flow's transmission offset of chips in the transmission path. The genetic algorithm is used to adjust the phase of scheduling tables of chips' send ports, whereby the flow's transmission offset can be updated and the maximum waiting time of the worst case can be shortened. The time-triggered scheduling table with global optimization significance can be obtained. Compared with the methods of generating TT scheduling tables using the SMT formalizer, such as Yices, this method does not work for a long time without stopping and can not cause undefined problems. The case study shows that the scheduler scale of this method increases by at least 30% for both symmetric and asymmetric structures. In addition, the ratio of the waiting time to the time period is taken as the normalized transmission delay metric. The case study also shows that the propagation delay value deduced by this method reduces to less than 2% of the one deduced by existing task-based scheduling methods.