Abstract:

Task scheduling in multicore systems becomes increasingly complex in the presence of inter-task dependencies, which traditional operating system schedulers are not designed to manage. As a result, tasks are often activated before their prerequisites are satisfied, leading to passive waiting and inefficient CPU utilization. While several dependency-aware approaches have been proposed in specialized domains such as heterogeneous or distributed computing, they typically require centralized control or static task graph knowledge, limiting their applicability to dynamic workloads. To address this limitation, the Dependency-Aware Model (DAM) was introduced as a lightweight user-space scheduling mechanism that defers task activation until all declared dependencies are resolved. This study presents a large-scale empirical evaluation of the DAM model using a simulation environment capable of executing compute-bound workloads under randomized configurations. The simulator supports parameterized variations in thread count, execution time, CPU affinity, and dependency density while ensuring the integrity of task graphs through enforced acyclicity. Over 39,000 simulation trials were conducted, capturing execution behavior under both standard and dependency-aware scheduling. The results demonstrate consistent performance improvements using DAM, with average execution time reductions exceeding 28% in single-test runs and 30% in ten-test batches. These findings confirm the model’s robustness and its effectiveness in reducing idle processor occupation across a wide range of scheduling scenarios.