/
RDMA Benchmarks Example for EPYC AMD
RDMA Benchmarks Example for EPYC AMD
Here is a simple example script to test RDMA verbs on two nodes.
1. Before you start make sure you understand the server topology (NUMA), for best performance select the closest NUMA to the adapter
See example here: HowTo Find the local NUMA node in AMD EYPC Servers
In this example, we see the the NUMA that is closest to the adapter is NUMA 2.
2. Get access to two nodes in the cluster. In case you are using slurm you can use salloc to allocate two nodes.
$ salloc -N 2 -p venus
3. Run manually the RDMA verbs scripts or use the following script, and bind the CPU NUMA to NUMA 2 (local).
TAG=$(date +%Y%m%d-%H%M%S) DEST=venus002 SRC=$HOSTNAME HCA="-d mlx5_0 -i 1" BINDS="numactl --cpunodebind=2 " BINDC="numactl --cpunodebind=2 " LTFLAGS="-a $HCA -F " BWFLAGS="-a $HCA --report_gbits -F " $BINDS ib_write_lat $LTFLAGS & ssh $DEST "$BINDC ib_write_lat $LTFLAGS $SRC " 2>&1 |tee log-lat-$TAG.txt $BINDS ib_write_bw $BWFLAGS & ssh $DEST "$BINDC ib_write_bw $BWFLAGS $SRC " 2>&1 |tee log-bw-$TAG.txt $BINDS ib_read_lat $LTFLAGS & ssh $DEST "$BINDC ib_read_lat $LTFLAGS $SRC " 2>&1 |tee log-lat-$TAG.txt $BINDS ib_read_bw $BWFLAGS & ssh $DEST "$BINDC ib_read_bw $BWFLAGS $SRC " 2>&1 |tee log-bw-$TAG.txt
Related content
Basic RDMA Benchmark Examples for AMD 2nd Gen EPYC CPUs over HDR InfiniBand
Basic RDMA Benchmark Examples for AMD 2nd Gen EPYC CPUs over HDR InfiniBand
More like this
RDMA Loopback test
RDMA Loopback test
More like this
HowTo Find the local NUMA node in AMD EPYC Servers
HowTo Find the local NUMA node in AMD EPYC Servers
More like this
OSU Benchmark Tuning for 2nd Gen AMD EPYC using HDR InfiniBand over HPC-X MPI
OSU Benchmark Tuning for 2nd Gen AMD EPYC using HDR InfiniBand over HPC-X MPI
More like this
GPUDirect Benchmarking
GPUDirect Benchmarking
More like this
Random Ring Effective Bandwidth (b_eff) with Adaptive Routing Analysis
Random Ring Effective Bandwidth (b_eff) with Adaptive Routing Analysis
Read with this