Tensity-Aware Optimized Scheduling of Parallel Real-Time Tasks on Multiprocessors

The federated scheduling framework is a popular multicore scheduling policy for parallel periodic real-time tasks that are often modeled as directed acyclic graphs (DAGs). However, it often over-estimates the processing requirements of parallel task execution, resulting in acute resource underutilization of available processing capacity in an already resource-constrained system. In this work, we aim to reduce resource under-utilization by proposing HL-DAGs, where compatible DAG tasks are fused and transformed into a fork-join DAG task model to opportunistically reclaim the usable utilization of the system. HL-DAGs, however, may fail to meet a task’s timing requirements and impact the schedulability of the system. To tackle this challenge, we present a technique to enforce both the logical and timing correctness requirements of a HL-DAG task. In addition, we discuss a fixed-priority partitioned scheduling algorithm (HL-FED) to schedule HL-DAGs, along with other DAG tasks, on multicore systems. Simulation results indicate that HL-FED can improve the usable system utilization by 27% on average, and up to 33%, over existing DAG scheduling frameworks. In addition, our proposed solution can also tighten the processing capacity by up to 11% when compared to the state-of-the-art federated scheduling framework.