Adaptive OpenMP Task Scheduling Using Runtime APIs and Machine Learning

Qawasmeh, Ahmad; Malik, Abid M.; Chapman, Barbara

doi:10.1109/icmla.2015.111

Cited by 5 publications

(3 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…All tasks in a parallel task set have to essentially start at the same time. Parallel for loops, a feature of numerous parallel programming languages, including OpenMP and CilkPlus, are used to create these tasks (Saifullah et al , 2013; Qawasmeh et al , 2016). The tasks can also be clustered based on certain weights or other characteristics of the tasks.…”

Section: Methodsmentioning

confidence: 99%

Utilization and criticality based fault-tolerant scheduling in multicore mixed critical systems

Godabole¹,

Bhole²

2023

IJPCC

View full text Add to dashboard Cite

Purpose The main purpose of the paper is timing analysis of mixed critical applications on the multicore system to identify an efficient task scheduling mechanism to achieve three main objectives improving schedulability, achieving reliability and minimizing the number of cores used. The rise in transient faults in embedded systems due to the use of low-cost processors has led to the use of fault-tolerant scheduling and mapping techniques. Design/methodology/approach The paper opted for a simulation-based study. The simulation of mixed critical applications, like air traffic control systems and synthetic workloads, is carried out using a litmus-real time testbed on an Ubuntu machine. The heuristic algorithms for task allocation based on utilization factors and task criticalities are proposed for partitioned approaches with multiple objectives. Findings Both partitioned earliest deadline first (EDF) with the utilization-based heuristic and EDF-virtual deadline (VD) with a criticality-based heuristic for allocation works well, as it schedules the air traffic system with a 98% success ratio (SR) using only three processor cores with transient faults being handled by the active backup of the tasks. With synthetic task loads, the proposed criticality-based heuristic works well with EDF-VD, as the SR is 94%. The validation of the proposed heuristic is done with a global and partitioned approach of scheduling, considering active backups to make the system reliable. There is an improvement in SR by 11% as compared to the global approach and a 17% improvement in comparison with the partitioned fixed-priority approach with only three processor cores being used. Research limitations/implications The simulations of mixed critical tasks are carried out on a real-time kernel based on Linux and are generalizable in Linux-based environments. Practical implications The rise in transient faults in embedded systems due to the use of low-cost processors has led to the use of fault-tolerant scheduling and mapping techniques. Originality/value This paper fulfills an identified need to have multi-objective task scheduling in a mixed critical system. The timing analysis helps to identify performance risks and assess alternative architectures used to achieve reliability in terms of transient faults.

show abstract

Section: Methodsmentioning

confidence: 99%

Utilization and criticality based fault-tolerant scheduling in multicore mixed critical systems

Godabole¹,

Bhole²

2023

IJPCC

View full text Add to dashboard Cite

show abstract

“…APARF [15] is an adaptative runtime framework to enhance peformance of OpenMP task-based programs. It makes use of PAPI to obtain performance events and trains an ANN to find which scheduling scheme should be used to obtain the optimal performance in unseen programs.…”

Section: Related Workmentioning

confidence: 99%

Predicting number of threads using balanced datasets for openMP regions

et al. 2022

View full text Add to dashboard Cite

Incorporating machine learning into automatic performance analysis and tuning tools is a promising path to tackle the increasing heterogeneity of current HPC applications. However, this introduces the need for generating balanced datasets of parallel applications’ executions and for dealing with natural imbalances for optimizing performance parameters. This work proposes a holistic approach that integrates a methodology for building balanced datasets of OpenMP code-region patterns and a way to use such datasets for tuning performance parameters. The methodology uses hardware performance counters to characterize the execution of a given region and correlation analysis to determine whether it covers an unique part of the pattern input space. Nevertheless, a balanced dataset of region patterns may become naturally imbalanced when used for training a model for tuning any specific performance parameter. For this reason, we have explored several methods for dealing with naturally imbalanced datasets for finding the appropriated way of using them for tuning purposes. Experimentation shows that the proposed methodology can be used to build balanced datasets and that such datasets, plus a combination of Random Forest and binary classification, can be used to train a model able to accurately tune the number of threads of OpenMP parallel regions.

show abstract

“…Qawasmeh et al [20] analyze timing and cache performance of runtime events including task creation to decide on optimal scheduling strategies in the OpenUH runtime system. They extend [21] the Sun/Oracle Collector API [9] to record the events.…”

Section: Related Workmentioning

confidence: 99%

Extending OMPT to Support Grain Graphs

Langdal

Jahre

Muddukrishna

2017

Scaling OpenMP for Exascale Performance and Portability

View full text Add to dashboard Cite

Abstract. The upcoming profiling API standard OMPT can describe almost all profiling events required to construct grain graphs, a recent visualization that simplifies OpenMP performance analysis. We propose OMPT extensions that provide the missing descriptions of task creation and parallel for-loop chunk scheduling events, making OMPT a sufficient, standard source for grain graphs. Our extensions adhere to OMPT design objectives and incur a low overhead for BOTS (up to 2% overhead) and SPEC OMP2012 (1%) programs. Although motivated by grain graphs, the events described by the extensions are general and can enable cost-effective, precise measurements in other profiling tools as well.

show abstract

Adaptive OpenMP Task Scheduling Using Runtime APIs and Machine Learning

Cited by 5 publications

References 23 publications

Utilization and criticality based fault-tolerant scheduling in multicore mixed critical systems

Utilization and criticality based fault-tolerant scheduling in multicore mixed critical systems

Predicting number of threads using balanced datasets for openMP regions

Extending OMPT to Support Grain Graphs

Contact Info

Product

Resources

About