​netperf

The netperf tool is a strong choice for both throughput and latency testing.

• The TCP_​STREAM test in netperf measures the throughput capabilities of the device.

  Server command example:

  netserver

  Client command example:

  netperf -t TCP_STREAM -l 30 -H <serverIP>

• The TCP_​RR test in netperf returns latency as transactions per second. This is a round-trip value. Use the -T x,x option, where x is a CPU local to the device.

  One-way latency can be calculated using:

  Latency (usec) = (0.5/[Transactions/sec])*1,000,000

  Server command example: netserver

  Client command example:

  netperf -t TCP_RR -l 30 -H <serverIP> -T x,x Example: netperf -t TCP_RR -H <serverIP> -T 8,12 8 ==> Pin the client-side netperf process to CPU 8 12 ==> Request the server-side netserver process to run on CPU 12

• To start multiple instances or threads of netperf, use a for-loop to map threads to TCP ports and run netperf in the background using& to create multiple parallel processes. Server command example, start 8 threads:

  Server command example, start 8 threads:

  for i in {0..7}; do netserver -L $serverIP -p 520$i & done

  Client command example, start 8 threads:

  for i in {0..7}; do netperf -H $serverIP -p 520$i -t TCP_STREAM -l 30 & done