High Performance MPI over the Slingshot Interconnect

The Slingshot interconnect designed by HPE/Cray is becoming more relevant in high-performance computing with its deployment on the upcoming exascale systems. In particular, it empowering first and highest-ranked supercomputer world, Frontier. It offers various features such as adaptive routing, congestion control, isolated workloads. of newer interconnects sparks interest related to performance, scalability, any potential bottlenecks they are critical elements contributing scalability across nodes these this paper, we delve into challenges poses current state-of-the-art MPI (message passing interface) libraries. look at performance when using nodes. We present a comprehensive evaluation communication libraries including Cray MPICH, Open- + UCX, RCCL, MVAPICH2 CPUs GPUs Spock system, an early access cluster deployed Slingshot-10, AMD MI100 Epyc Rome emulate Frontier system. also evaluate preliminary CPU-based support Slingshot-11 interconnect.