Timing-Anomaly Free Dynamic Scheduling of Conditional DAG Tasks on Multi-Core Systems

Abstract

In this paper, we propose a novel approach to schedule conditional DAG parallel tasks, with which we can derive safe response time upper bounds significantly better than the state-of-the-art counterparts. The main idea is to eliminate the notorious timing anomaly in scheduling parallel tasks by enforcing certain order constraints among the vertices, and thus the response time bound can be accurately predicted off-line by somehow “simulating” the runtime scheduling. A key challenge to apply the timing-anomaly free scheduling approach to conditional DAG parallel tasks is that at runtime it may generate exponentially many instances from a conditional DAG structure. To deal with this problem, we develop effective abstractions, based on which a safe response time upper bound is computed in polynomial time. We also develop algorithms to explore the vertex orders to shorten the response time bound. The effectiveness of the proposed approach is evaluated by experiments with randomly generated DAG tasks with different parameter configurations.

A Global DAG Task Scheduler Using Deep Reinforcement Learning and Graph Convolution Network

“…In the real-time system literature, fixed-priority scheduling schemes for parallel tasks have been receiving attention relatively recently in [1], [2], while many works on DAG task scheduling considered dynamic scheduling cases [3], [5], [6], [12], [13]. By fixed-priority scheduling, subtasks (or nodes) in a DAG task can be scheduled globally on all processors, and furthermore, a platform can be free from overhead issues of online scheduling in real-time systems [1], [14].…”

Bounding the execution time of parallel applications on unrelated multiprocessors

“…The Cilk-based parallel applications are modeled with a restricted version of DAGs, and the bound is extended to cover the work-stealing scheduler. In Voudouris et al (2017), Chen et al (2019) a formally proven timing anomaly-free dynamic scheduler is introduced that provides tighter and more scalable, for the number of tasks and number of processors, makespan estimations. The results of Voudouris et al (2017), Chen et al (2019) cannot be trivially applied to unrelated multiprocessors because the DAG can have different schedule lengths depending on the task-processor mapping.…”

“…One of the fundamental problems is the presence of timing anomalies (Graham 1969). Note that a timing anomaly is already known to exist for the homogeneous multiprocessor model, which is a special case of the unrelated multiprocessor model (Voudouris et al 2017;Pathan et al 2018;Chen et al 2019). Therefore, an example of a DAG-similar to that of (Voudouris et al 2017) can also be constructed to demonstrate the presence of timing anomalies in the unrelated multiprocessor model.…”

“…Therefore, an example of a DAG-similar to that of (Voudouris et al 2017) can also be constructed to demonstrate the presence of timing anomalies in the unrelated multiprocessor model. A method to avoid such anomalies for homogeneous multiprocessors is to preserve strictly, also at run-time, the order of start time of the execution of the tasks that was determined at analysis time (Voudouris et al 2017;Pathan et al 2018;Chen et al 2019). Unfortunately, enforcing such an order of starting the tasks' execution is not enough to guarantee anomaly-freedom on unrelated machines because of the different speed relationships that each task has with each processor type.…”

Partial order based non-preemptive communication scheduling towards real-time networks-on-chip

“…The recent work [9] has addressed the timing anomalies for multi-cores, but their approach cannot be trivially extended from multi-core to NoC scheduling due to direct and indirect contention in communication resources, showing pessimistic runtime performance. Besides, when considering the joint scheduling of computation and communication tasks, all of the possible runtime execution scenarios can be exponential.…”