Argonne National Laboratory

Malleable Model Coupling with Prediction

TitleMalleable Model Coupling with Prediction
Publication TypeConference Paper
Year of Publication2012
AuthorsKim, D-H, Larson, JW, Chiu, K
Conference NameProceedings of the The 12th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGRID 2012)
Date Published05/2012
PublisherProceedings of the The 12th IEEE/AC
Conference LocationOttawa, Canada
Other NumbersANL/MCS-P1984-1211

Achieving ultrascalability in coupled multiphysics and multiscale models requires dynamic load balancing both within and between their constituent subsystems. Interconstituent dynamic load balance requires runtime resizing or malleability of subsystem PE cohorts. We enhance the Malleable Model Coupling Toolkits Load Balance Manager (LBM) to incorporate prediction of a coupled systems constituent computation times and coupled model global iteration time. The prediction system employs piecewise linear and cubic spline interpolation of timing measurements to guide constituent cohort resizing. Performance studies of the new LBM using a simplified coupled model testbed similar to a coupled climate model show dramatic improvement (77%) in the LBMs convergence rate.