A parallel domain decomposition algorithm for coastal ocean circulation models based on integer linear programming
- 133 Downloads
This paper presents a new parallel domain decomposition algorithm based on integer linear programming (ILP), a mathematical optimization method. To minimize the computation time of coastal ocean circulation models, the ILP decomposition algorithm divides the global domain in local domains with balanced work load according to the number of processors and avoids computations over as many as land grid cells as possible. In addition, it maintains the use of logically rectangular local domains and achieves the exact same results as traditional domain decomposition algorithms (such as Cartesian decomposition). However, the ILP decomposition algorithm may not converge to an exact solution for relatively large domains. To overcome this problem, we developed two ILP decomposition formulations. The first one (complete formulation) has no additional restriction, although it is impractical for large global domains. The second one (feasible) imposes local domains with the same dimensions and looks for the feasibility of such decomposition, which allows much larger global domains. Parallel performance of both ILP formulations is compared to a base Cartesian decomposition by simulating two cases with the newly created parallel version of the Stevens Institute of Technology’s Estuarine and Coastal Ocean Model (sECOM). Simulations with the ILP formulations run always faster than the ones with the base decomposition, and the complete formulation is better than the feasible one when it is applicable. In addition, parallel efficiency with the ILP decomposition may be greater than one.
KeywordsCoastal ocean circulation model Parallel computing Domain decomposition Integer linear programming Urban ocean ECOM
This work was partially funded by a research task agreement entered between the Trustees of the Stevens Institute of Technology and the Port Authority of New York and New Jersey, effective August 19, 2014. The authors would like to acknowledge the unwavering support of Mr. David Dodd, Vice President for Information Technology and CIO at Stevens Institute of Technology. We would also like to thank Karen Swift, Kevin Ying, Joe Formoso, David Runnels, Kurt Hockenbury, and James Cunningham, for their assistance with the Pharos Hyperscale Supercomputing Facility.
- Dennis JM (2007) Inverse space-filling curve partitioning of a global ocean model. Parallel and Distributed Processing Symposium, 2007 IPDPS 2007 I.E. International, IEEE, pp 1–10Google Scholar
- Dunning I, Huchette J and Lubin M (2015) Jump: a modeling language for mathematical optimization. arXiv preprint arXiv:150801982Google Scholar
- Georgas N and Blumberg AF (2010) Establishing confidence in marine forecast systems: the design and skill assessment of the New York Harbor observation and prediction system, version 3 (NYHOPS v3). 11th International Conference in Estuarine and Coastal Modeling (ECM11) Spalding, ML, Ph D, PE, American Society of Civil Engineers, Washington, pp 660–685Google Scholar
- Georgas N, Blumberg AF, Herrington TO (2007) An operational coastal wave forecasting model for New Jersey and Long Island waters. Shore and Beach 75:30Google Scholar
- Georgas N, Blumberg AF, Bruno MS and Runnels DS (2009) Marine forecasting for the New York urban waters and harbor approaches: the design and automation of NYHOPS. 3rd International Conference on Experiments/Process/System Modelling/Simulation & Optimization Demos T Tsahalis, Ph D, University of Patras, Greece, pp 345–352Google Scholar
- Leung JY (2004) Handbook of scheduling: algorithms, models, and performance analysis. CRC Press, Boca RatonGoogle Scholar
- Orton P, Georgas N, Blumberg A, Pullen J (2012) Detailed modeling of recent severe storm tides in estuaries of the New York city region. J Geophys Res Oceans (1978–2012) 117Google Scholar
- Sierksma G (2001) Linear and integer programming: theory and practice. CRC Press, Boca RatonGoogle Scholar