The topology information gives us a rough understanding of the expected performance. We complement this with real measurements conducted on the hardware. For this purpose, we use ''pairwise'' a micro benchmark that ping-pongs messages between any combination of cores.
The benchmark measures the send, receive and roundtrip times, i.e. the time it takes until smlt_qp_send() or smlt_qp_recv() return.
All information and benchmarks of machine appenzeller can be found here.
All information and benchmarks of machine babybel can be found here.
All information and benchmarks of machine gottardo can be found here.
All information and benchmarks of machine gottardo-ht can be found here.
All information and benchmarks of machine gruyere can be found here.
All information and benchmarks of machine nos can be found here.
All information and benchmarks of machine phi can be found here.
All information and benchmarks of machine pluton can be found here.
All information and benchmarks of machine sbrinz can be found here.
All information and benchmarks of machine sgs-r815-03 can be found here.
All information and benchmarks of machine sgs-r820-01 can be found here.
All information and benchmarks of machine smaug-2 can be found here.
All information and benchmarks of machine smaug-3 can be found here.
All information and benchmarks of machine tomme can be found here.
All information and benchmarks of machine ziger can be found here.