High Performance & Scalable MPI Library Over Broadcom RoCE
Monday, May 13, 2024 3:00 PM to Wednesday, May 15, 2024 4:00 PM · 2 days 1 hr. (Europe/Berlin)
Foyer D-G - 2nd floor
Research Poster
Interconnects and NetworksParallel Programming Languages
Information
Poster is on display and will be presented at the poster pitch session.
RDMA over converged Ethernet (RoCE) allows remote direct memory access (RDMA) over an Ethernet network by encapsulating Infiniband (IB) transport packet over Ethernet. Broadcom Ethernet network adapters support RoCE as a complete hardware offload feature, which provides direct memory access for applications bypassing the CPU. In this poster, we first characterize the RDMA performance of Broadcom Thor RoCEv2 adapter by measuring MPI level overheads compared to IB level send/recv performance. Then, we introduce how we Optimize the MVAPICH2 MPI library with its native support of ibverbs interface on systems configured with Broadcom Thor adapter. To validate the efficacy of our work, we also evaluate Micro-benchmark and Application-level performance to validate MPI optimizations.
Contributors:
RDMA over converged Ethernet (RoCE) allows remote direct memory access (RDMA) over an Ethernet network by encapsulating Infiniband (IB) transport packet over Ethernet. Broadcom Ethernet network adapters support RoCE as a complete hardware offload feature, which provides direct memory access for applications bypassing the CPU. In this poster, we first characterize the RDMA performance of Broadcom Thor RoCEv2 adapter by measuring MPI level overheads compared to IB level send/recv performance. Then, we introduce how we Optimize the MVAPICH2 MPI library with its native support of ibverbs interface on systems configured with Broadcom Thor adapter. To validate the efficacy of our work, we also evaluate Micro-benchmark and Application-level performance to validate MPI optimizations.
Contributors:
Format
On-site
