TY - GEN
T1 - OpenSHMEM Active Message Extension for Task-Based Programming
AU - Lu, Wenbin
AU - Curtis, Tony
AU - Chapman, Barbara
N1 - Publisher Copyright:
© 2022, Springer Nature Switzerland AG.
PY - 2022
Y1 - 2022
N2 - As a lightweight library-based Partitioned Global Address Space (PGAS) programming model, OpenSHMEM provides efficient one-sided and collective communications and is receiving more attention in recent years. However, task-based programming models are getting bigger traction in scientific computing communities. Application developers are attracted by their ability to achieve better load balance in the face of ever-growing application complexity, and the increasing on-node parallelism in modern high-performance computing machines. Although communication contexts provide threads with first-class access to the network in the OpenSHMEM+X model, OpenSHMEM still has very limited ability to perform advanced operations found in other task-based models. For example, compared to the remote procedure call (RPC) mechanism in the UPC++ programming model, more work is required if the signal/wait routines are used to achieve similar remote task launching operations. In this paper, we introduce a lightweight active message (AM) extension to OpenSHMEM that is designed to perform short, non-blocking remote function invocations. This extension aims to bring some benefits of task-based programming to OpenSHMEM without making it a full-blown heavyweight tasking system with a sophisticated scheduler. We study the performance of this active message extension by running micro-benchmarks, and by evaluating its computation efficiency at different task granularities using the TaskBench framework.
AB - As a lightweight library-based Partitioned Global Address Space (PGAS) programming model, OpenSHMEM provides efficient one-sided and collective communications and is receiving more attention in recent years. However, task-based programming models are getting bigger traction in scientific computing communities. Application developers are attracted by their ability to achieve better load balance in the face of ever-growing application complexity, and the increasing on-node parallelism in modern high-performance computing machines. Although communication contexts provide threads with first-class access to the network in the OpenSHMEM+X model, OpenSHMEM still has very limited ability to perform advanced operations found in other task-based models. For example, compared to the remote procedure call (RPC) mechanism in the UPC++ programming model, more work is required if the signal/wait routines are used to achieve similar remote task launching operations. In this paper, we introduce a lightweight active message (AM) extension to OpenSHMEM that is designed to perform short, non-blocking remote function invocations. This extension aims to bring some benefits of task-based programming to OpenSHMEM without making it a full-blown heavyweight tasking system with a sophisticated scheduler. We study the performance of this active message extension by running micro-benchmarks, and by evaluating its computation efficiency at different task granularities using the TaskBench framework.
KW - Active message
KW - OpenSHMEM
KW - PGAS
KW - Tasking
UR - https://www.scopus.com/pages/publications/85130964226
U2 - 10.1007/978-3-031-04888-3_8
DO - 10.1007/978-3-031-04888-3_8
M3 - Conference contribution
AN - SCOPUS:85130964226
SN - 9783031048876
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 129
EP - 143
BT - OpenSHMEM and Related Technologies. OpenSHMEM in the Era of Exascale and Smart Networks - 8th Workshop on OpenSHMEM and Related Technologies, OpenSHMEM 2021, Revised Selected Papers
A2 - Poole, Stephen
A2 - Hernandez, Oscar
A2 - Baker, Matthew
A2 - Curtis, Tony
PB - Springer Science and Business Media Deutschland GmbH
T2 - 8th Workshop on OpenSHMEM and Related Technologies, OpenSHMEM 2021
Y2 - 14 September 2021 through 16 September 2021
ER -