The command to start traffic takes the following form
start PORT MODE [OPTIONS...]
You can generate three types of traffic
udp: Simple stream(s) of UDP packetsflowgen: More realistic, flow-based traffic
Here are a few useful commands to get started quickly. See Options below for more a more detailed configurations.
A simple benchmarking scenario might involve measuring how well a DUT copes with
various amounts of incoming flows. You can generate 10,000 infinitely long-lived
UDP flows containing 60 byte packets with the following command. You can
identify which port to generate traffic on by either its PCI address in
BUS:SLOT.FUNCTION format, or by its integral DPDK port id.
localhost:10514 $ add port portA pci='03:00.0'
localhost:10514 $ start portA udp num_flows=10000, pkt_size=60
You can then check how well the DUT is keeping up by running monitor port
which will produce the output below. The INC columns indicate the rate of
return traffic from the DUT along with the average, median and 99th
percentile/jitter round-trip times experienced by packets (reported in
microseconds). monitor port also dumps a comma separated form of its output
to /tmp/bench.csv. You can change where stats get dumped for future runs by
running set csv <PAHT_TO_YOUR_CSV>.
localhost:10514 $ monitor port
Monitoring ports: 03:00.0 (Send CTRL + c to stop)
14:40:49.469716 INC Mbps Mpps dropped avg_rtt (us) med_rtt (us) 99_rtt (us) avg_jit (us) med_jit (us) 99_jit (us) OUT Mbps Mpps dropped
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
03:00.0/PMDPort 10128.9 15.070 3582 15.314 14.650 18.200 0.822 0.650 2.100 10134.5 15.081 143
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Running show config will dump the current configuration of all active ports,
including information like MAC addresses, a list of cores used, and other
mode-specific settings:
localhost:10514 $ show config
Port 03:00.0
-----------
rfc2544_loss_rate: disabled
rfc2544_window: 5
rfc2544_warmup: 2
rfc2544_drain: 5
rfc2544_adj: 10
rfc2544_max_rounds: 10
pps: <= line rate
mbps: <= line rate
src_mac: 3c:fd:fe:a2:a6:e0
dst_mac: 02:1e:67:9f:4d:bb
src_ip: 192.168.0.1
dst_ip: 10.0.0.1
tx_cores: 0
rx_cores: 0
pkt_size: 60
num_flows: 1000
imix: disabled
-----------
To stop generating traffic just enter stop 03:00.0 at the prompt.
If you're benchmarking a DUT over a fast NIC, you may need to use multiple cores
to saturate the link. You can do that by appending the cores option to your
command like so:
localhost:10514 $ add port portA pci='03:00.0'
localhost:10514 $ start portA udp num_flows=10000, pkt_size=60, cores="0 1"
Run monitor ports again and note the higher rates in the OUT columns.
localhost:10514 $ monitor port
Monitoring ports: 03:00.0 (Send CTRL + c to stop)
14:40:49.469716 INC Mbps Mpps dropped avg_rtt (us) med_rtt (us) 99_rtt (us) OUT Mbps Mpps dropped
---------------------------------------------------------------------------------------------------------------------------------------------
03:00.0/PMDPort 12551.5 18.124 775004 38.612 28.775 65.300 21692.3 32.280 782
---------------------------------------------------------------------------------------------------------------------------------------------
There are several knobs you can tweak to control traffic. They are specified
like foo="bar", baz=3.
The following options are applicable to all modes of traffic.
| Option | Description |
|---|---|
pps |
Setting pps=X will result in trafficgen sending traffic at X packets per second. When set with rfc2544_loss_rate, trafficgen will start sending at X packets per second before adjusting the sending rate. By default pps is not set, meaning trafficgen will try to send at line rate. |
mbps |
Setting mbps=X will result in trafficgen sending traffic at X megabits per second. By default mbps is not set and trafficgen will try to send at line rate. |
tx_cores, rx_cores |
Setting tx_cores and rx_cores will split the work of sending and receiving packets on a port across multiple cores. For example, setting tx_cores="0 1" and rx_cores="2 3" will result in trafficgen splitting traffic sending packets on cores 0 and 1 and receiving them on cores 2 and 3. By default {t,r}x_cores are not set and trafficgen will send and receive on the same core, chosen at random. |
src_mac, dst_mac |
Setting src_mac=X and dst_mac=Y will result in trafficgen setting the source and destination Ethernet addresses to X and Y respectively. By default src_mac is set to the Ethernet address of the sending port and dst_port is set to 02:1e:67:9f:4d:bb. |
src_ip, dst_ip |
Setting src_ip=X and dst_ip=Y will result in trafficgen generating flows with source and destination IP addresses starting at X and Y respectively. By default src_ip and dst_ip are 192.168.0.1 and 10.0.0.1 respectively. |
tx_timestamp_offset, rx_timestamp_offset |
The offset in bytes to place and read timestamps for latency measurements. By default tx_timestamp_offset and rx_timestamp_offset are unset and trafficgen will stamp and read timestamps from the first 8 bytes of the TCP/UDP payload. However, in some cases, when the DUT performs encapsulation or decapsulation or your mode's pipeline uses header options, you may need to change these offsets to see meaningful latency measurements. Setting {r,t}x_timestamp_offset=N will result in trafficgen stamping and reading timestamps N bytes from the beginning of the packet (e.g., N=42 will place timestamps at the beginning of a UDP payload. 14 bytes for the Ethernet header, 20 for the IP header and 8 for the UDP header). |
rfc2544_loss_rate |
Setting rfc2544_loss_rate=X will result in trafficgen sending traffic at rate such that it sees a frame last rate of X (in [0,1]) as defined by RFC2544. If rfc2544_loss_rate is set, pps must be set to the initial sending rate. By befault rfc2544_loss_rate is not set and trafficgen sends at whatever rate you tell it to. |
rfc2544_window |
Setting rfc2544_window=X will result in the RFC 2544 process for this port using X second rounds excluding the warmup and draining periods. The default round duration is 30 seconds. |
rfc2544_warmup |
Setting rfc2544_warmup=X will result in the RFC 2544 process for this port ignoring the first X seconds ,i.e. not take any measurements, for each round. The default is warmup duration is 5 seconds. |
rfc2544_drain |
Setting rfc2544_drain=X will result in the RFC 2544 process for this port waiting for X seconds for its RX queues to drain before beginning a new round. The default queue drainage duration is 5 seconds. |
rfc2544_adj |
Setting rfc2544_adj=X will result in the RFC 2544 process for this port dropping its sending rate by X% after any round it experiences more than loss_rate% packet loss. After experiencing packet loss <= loss_rate% for two consecutive rounds it will increase its sending rate by X%. |
rfc2544_max_rounds |
Setting rfc2544_max_rounds=N will limit the RFC 2544 process for this port to N rounds of searching for the ideal sending rate. The default is 10 rounds. |
The following options are only applicable to ports sending in udp mode.
| Option | Description |
|---|---|
num_flows |
Setting num_flows=N will result in trafficgen sending N flows. By default num_flows is set to 10. |
pkt_size |
Setting pkt_size=X will result in trafficgen sending X-byte packets. By default pkt_size is set to 60. |
imix |
Setting imix=1 will result in trafficgen generate Internet Mix-like traffic. By default imix is not set and trafficgen generates fixed-size packets of pkt_size bytes. |
The following options are only applicable to ports sending in flowgen mode.
| Option | Description |
|---|---|
pkt_size |
Setting pkt_size=X will result in trafficgen sending X-byte packets. By default pkt_size is set to 60. |
num_flows |
Setting num_flows=N will result in trafficgen sending N flows. By default num_flows is set to 10. |
flow_duration |
Setting flow_duration=X will result in trafficgen generating flows that live for up to X seconds. By default flow_duration is set to 5 seconds. |
flow_rate |
Setting flow_rate=R will result in trafficgen generating R new flows per second. By default flow_rate is set to num_flows / flow_duration. |
arrival |
Setting arrival will change the distribution from which trafficgen generates flow IDs (5-tuples). It can be either pareto or uniform. By default arrival is set to uniform. |
duration |
Setting duration will change the distribution from which trafficgen chooses flow lifetimes. It can be either pareto or uniform. By default duration is set to uniform. |