OHIO: Improving RDMA Network Scalability in MPI_Alltoall Through Optimized Communication Design
HOTI - Hot Interconnects Symposium via YouTube
Overview
Watch a 26-minute technical conference presentation from the Hot Interconnects Symposium that explores optimizing RDMA network scalability for MPI_Alltoall operations through hierarchical and intra/inter-node communication overlap design. Learn from researchers Tu Tran, Goutham Kalikrishna Reddy Kuncham, Bharath Ramesh, Shulei Xu, Hari Subramoni, Mustafa Abduljabbar and Dhabaleswar K. Panda as they present their technical findings on improving communication efficiency in distributed systems. Gain insights into advanced optimization techniques for collective operations in high-performance computing environments, presented as part of Technical Paper Session D chaired by Craig Stunkel from Nvidia.
Syllabus
Day 2 13:00: OHIO: Improving RDMA Network Scalability in MPI_Alltoall
Taught by
HOTI - Hot Interconnects Symposium