Performance Modeling and Tuning Strategies of Mixed Mode Collective Communications Session: Collective and Group Communication Event Type: Paper Time: 10:30am - 11:00am Session Chair: Dhabaleswar K. (DK) Panda Speaker(s): Meng-Shiou Wu, Ricky A. Kendall, Zhao Zhang, Kyle Wright Location: 608-609 Abstract: On SMP clusters, mixed mode collective MPI communications, which use shared memory communications within SMP nodes and point-to-point communications between SMP nodes, are more efficient than conventional implementations. In a previous study, we proposed several new methods that made mixed mode collective communications significantly faster than the pure point-to-point ones. However, the optimal performance required the tuning of many parameters, which was done by testing every possible setting and was very time consuming. In this study, we propose a new performance model that considers the special characteristics of mixed mode collective communications. The model provides good predictions to reduce most settings without testing by execution. It considers both shared-memory and point-to-point communications, while existing performance models only consider the point-to-point ones. Based on this model, we develop a number of tuning strategies that reduce the tuning time to only 10% of previous tuning time. This paper can be found in the ACM and IEEE Digital Libaries Click here for ACM Click here for IEEE Chair/Speaker Details: Dhabaleswar K. (DK) Panda (Chair) The Ohio State University Meng-Shiou Wu Scalable Computing Laboratory, Ames Laboratory, U.S. DOE/ Department of Electrical and Computer Engineering, Iowa State University Ricky A. Kendall Scalable Computing Laboratory, Ames Laboratory, U.S. DOE/Department of Computer Science, Iowa State University Zhao Zhang Department of Electrical and Computer Engineering, Iowa State University Kyle Wright Scalable Computing Laboratory, Ames Laboratory, U.S. DOE/Department of Computer Science, Iowa State University