shenango-bsd/apps/dpdk_netperf

Latency Benchmarks

First build DPDK (without driver modifications), then build dpdk_netperf in this directory with make clean && make.

DPDK only

To run the benchmark with pure DPDK on both machines:

On the server (IP 192.168.1.2):

sudo ./build/dpdk_netperf -l2 --socket-mem=128 -- UDP_SERVER 192.168.1.2

On the client (IP 192.168.1.3):

sudo ./build/dpdk_netperf -l2 --socket-mem=128 -- UDP_CLIENT 192.168.1.3 192.168.1.2 50000 8001 10 8

Shenango spinning (IOKernel + runtime)

To run Shenango with the server runtime thread spinning, start the IOKernel and then in shenango/apps/bench:

./netbench_udp tbench.config server

Then run the client as above.

Shenango waking (IOKernel + runtime + wakeup)

To run with Shenango in its default mode but no batch work, start the IOKernel and then in shenango/apps/bench:

./netbench_udp waking.config server

Then run the client as above.

Shenango preempting (IOKernel + runtime + wakeup + preemption)

To run Shenango with a batch application running concurrently, start the IOKernel and then in shenango/apps/bench:

./stress stress.config 100 100 sqrt
./netbench_udp waking.config server

Then run the client as above. If your server does not have 24 hyperthreads, you will need to adjust runtime_kthreads in stress.config to be 2 fewer than the number of hyperthreads on your server.