High Performance RDMA-Based MPI Implementation over InfiniBand

By Jiuxing Liu, Jiesheng Wu, Dhabaleswar K. D. K. Panda. International Conference on Supercomputing, 2003

475 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

A generic communication scheduler for distributed DNN training acceleration (2019, methodology): “Existing approaches include: (1) speeding up individual messages by using RDMA [25] or NCCL [4]; (2) compressing data transmission such as gradient quantization [9, 37] and sparse parameter synchronization [7]; (3) optimizing communication approach, e.”
Kokkos: Enabling manycore performance portability through polymorphic memory access patterns (2014, unknown): “‘‘Raw’’ array-of-pointer variables are now incrementally replaced with View variables.”
Using RDMA efficiently for key-value services (2014, background): “While [30, 13, 20, 16, 19] have benchmarked verbs performance before, it has been for large messages in the context of applications like NFS and MPI.”
Datacenter RPCs can be General and Fast (2018, methodology): “This design is widely used in other systems such as NFS’s RPCs [20] and some MPI implementations [48].”
FaRM: Fast Remote Memory (2014, unknown): “In large clusters, it may be better to use RDMA write with immediate and a shared receive queue [35], which would make polling overhead constant.”