​Low Latency

• ​​Turn hyper-threading (logical cores) OFF.
• ​​Ensure the network device is local to numa core 0.
• ​​Pin the benchmark to core 0 using taskset -c 0.
• ​​Turn irqbalance off using systemctl stop irqbalance or systemctl disable irqbalance
• ​​Run the affinity script to spread across cores. Try either local or all.
• ​​Turn off interrupt moderation. ethtool -C ethX rx-usecs 0 tx-usecs 0 adaptive-rx off adaptive-tx off rx-usecs- high 0
• Limit number of queues to be equal to the number of cores on the local socket (32 in this example). ethtool -L ethX combined 32
• Pin interrupts to local socket only (script packaged with i40e driver source). set_irq_affinity -X local ethX
• ​​Use an established benchmark like netperf -t TCP_​RR, netperf -t UDP_​RR, or NetPipe. netperf -t TCP_RR or netperf -t UDP_RR
• Pin benchmark to a single core in the local NUMA node. taskset -c <cpu>