Skip to content

Archive for

RDTBench – Testing vSAN, RDMA and TCP between hosts

A while back I was asking engineering how they tested RDMA between hosts and stumbled upon RDTBench. This is a traffic generator, where you configure one host to act as a “server” and 1 to several hosts to act as clients communicating with it. This is a great tool for testing networking throughput before production use of a host, as well as validating RDMA configurations as it can be configured to generate vSAN RDMA traffic. Pings and IPERF are great, but being able to simulate RDT (vSAN protocol) traffic has it’s advantages.

RDT Bench does manifest itself as traffic on the vSAN Performance service host networking graphs
A few quick questions about it:

Where is it?

/usr/lib/vmware/vsan/bin

How do I run it?
You need to run it on two different hosts. One host will need to be configured to act as a client (by default it runs as a server). For the server I commonly use the -b flag to make it run bi-bidirectionally on the transport. -p rdma will run it in RDMA mode to test RDMA.

If RDMA is not working, go ahead and turn vSAN RDMA on (cluster, config, networking).

vSAN will “fail safe” back to TCP, but tell you what is missing from your configuration.

./rdtbench -h will provide the full list of command help.

For now this tool primarily exists for engineering (used for RDMA NIC validation) as well as support (as a more realistic alternative to IPERF), but I’m curious how we can incorporate it into other workflows for testing the health of a cluster.