Self-driving systems require multi-/many-core platforms with high computing power and low power consumption. However, for hard real-time applications, multiple demands on shared resources can impede real-time performance. Therefore, making the timing of memory access deterministic is important. The logical execution time (LET) paradigm has gained attention as a means to achieve this purpose. However, this approach lacks scalability owing to the overhead caused because the LET paradigm is set longer than the actual execution time of the task. This paper proposes a theoretical scheduling method for a model applying the LET paradigm to the directed acyclic graph (DAG) nodes for a multi-/many-core platform. The proposed method considers communication timing and generates a schedule that does not cause communication contentions. In addition, the proposed method performs a parallel calculation of tasks to deal with the overhead caused by adopting the LET paradigm.

Embedded systems, e.g., self-driving systems and advanced driver-assistance systems (ADAS), require computing platforms with high computing power and low power consumption. Multi-/many-core platforms satisfy these requirements effectively. However, for hard real-time applications, multiple demands on shared resources can impede real-time performance, and memory is one resource that can impair the desired performance significantly. Therefore, it is important that memory access timing be deterministic to facilitate predictability. To realize this, the Logical Execution Time (LET) paradigm is currently attracting attention. This paper proposes a theoretical scheduling method for a model applying the LET paradigm to directed acyclic graph (DAG) nodes for a multi-/many-core platform. The proposed method considers communication timing between nodes and generates a schedule that does not cause communication contentions. In addition, the proposed method attempts to distribute tasks and reduce LET intervals to address increased execution times due to the implementation of the LET paradigm. In the evaluation, we observed that the proposed method improved the schedule length by up to 40%.

Embedded systems such as self-driving systems require a computing platform with high computing power and low power consumption. Multi-/many-core platforms satisfy exactly these requirements. However, for hard real-time applications, multiple demands on shared resources can hinder real-time performance. Memory is among the resources that can most dramatically impair the desired performance. Therefore, we addressed contentions induced by the shared memory. We improve the predictability of contentions by dividing tasks into the memory access phase and the execution phase using a Directed Acyclic Graph (DAG). Existing methods are able to make accurate contention estimations for one Compute Cluster (CC) of a Clustered many-core processor. Our method is able to do the same for multiple CCs, thereby doubling the scalability in consideration of contentions. Using an Integer Linear Programming (ILP) formulation, we produced a static, non-preemptive, partitioned, time-triggered schedule. We also conducted an experiment in order to minimize the makespan. The evaluation confirmed that our new method reduced the makespan by increasing the number of CCs.

Computing platforms for embedded systems are increasingly being transformed into multi/many-core platforms because embedded systems have become extensive, complex, and automated. In the case of an autonomous driving system, various applications are simultaneously running, and low power consumption and large-scale calculation are required. Many-core processors with a multiple instruction, multiple data (MIMD) architecture can meet these requirements. This paper proposes a scheduling algorithm for an automotive driving system expressed in a directed acyclic graph (DAG) and we use Kalray MPPA-256 as the target many-core processor. On the basis of the architecture of Kalray MPPA-256, task processing that requires large-scale calculation and intercore communication is performed while avoiding communication contention by using a proposed grouping computational resource. In addition, we propose a scheduling method for a multi-rate DAG which is a DAG with multiple periods. This method generates a DAG task in a hyperperiod and schedules the DAG with dependency on tasks that have been released closely. The formulas for prioritization and processor selection are proposed for various generated tasks in a hyperperiod. Evaluation results show that the proposed algorithm is superior to existing DAG scheduling algorithms with regard to schedulability and deadline miss ratio.

Embedded systems such as self-driving systems and advanced driver-assistance systems (ADAS) require a computing platform with high computing power and low power consumption. Many-core platforms satisfy exactly these requirements. However, for hard real-time applications, multiple demands on shared resources can hinder real-time performance. Memory is among the resources that can most dramatically impair desired performance. In particular, it is important that the timing for accessing the memory is deterministic. As a method for realizing this, the Logical Execution Time (LET) paradigm is currently attracting attention. Research on the LET model is mainly conducted for control systems such as engines. However, scalability is required for application to autonomous driving systems, and high-speed task scheduling and mapping to the core must be considered. This paper summarizes the scheduling issues of hard real-time applications on a many-core processor using the LET model and discusses possible approaches.