TY - GEN
T1 - Machine-learning-based load balancing for community ice code component in CESM
AU - Balaprakash, Prasanna
AU - Alexeev, Yuri
AU - Mickelson, Sheri A.
AU - Leyffer, Sven
AU - Jacob, Robert
AU - Craig, Anthony
N1 - Publisher Copyright:
© Springer International Publishing Switzerland 2015.
PY - 2015
Y1 - 2015
N2 - Load balancing scientific codes on massively parallel architectures is becoming an increasingly challenging task. In this paper, we focus on the Community Earth System Model, a widely used climate modeling code. It comprises six components each of which exhibits different scalability patterns. Previously, an analytical performance model has been used to find optimal load-balancing parameter configurations for each component. Nevertheless, for the Community Ice Code component, the analytical performance model is too restrictive to capture its scalability patterns. We therefore developed machine-learning-based load-balancing algorithm. It involves fitting a surrogate model to a small number of load-balancing configurations and their corresponding runtimes. This model is then used to find high-quality parameter configurations. Compared with the current practice of expert-knowledge-based enumeration over feasible configurations, the machine-learning-based load-balancing algorithm requires six times fewer evaluations to find the optimal configuration.
AB - Load balancing scientific codes on massively parallel architectures is becoming an increasingly challenging task. In this paper, we focus on the Community Earth System Model, a widely used climate modeling code. It comprises six components each of which exhibits different scalability patterns. Previously, an analytical performance model has been used to find optimal load-balancing parameter configurations for each component. Nevertheless, for the Community Ice Code component, the analytical performance model is too restrictive to capture its scalability patterns. We therefore developed machine-learning-based load-balancing algorithm. It involves fitting a surrogate model to a small number of load-balancing configurations and their corresponding runtimes. This model is then used to find high-quality parameter configurations. Compared with the current practice of expert-knowledge-based enumeration over feasible configurations, the machine-learning-based load-balancing algorithm requires six times fewer evaluations to find the optimal configuration.
UR - https://www.scopus.com/pages/publications/84942593073
U2 - 10.1007/978-3-319-17353-5_7
DO - 10.1007/978-3-319-17353-5_7
M3 - Conference contribution
AN - SCOPUS:84942593073
SN - 9783319173528
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 79
EP - 91
BT - High Performance Computing for Computational Science - VECPAR 2014 - 11th International Conference, Revised Selected Papers
A2 - Marques, Osni
A2 - Dayde, Michel
A2 - Nakajima, Kengo
PB - Springer Verlag
T2 - 11th International Conference on High Performance Computing for Computational Science, VECPAR 2014
Y2 - 30 June 2014 through 3 July 2014
ER -