193. VF L3 forwarding kernel PF test in LPM mode with IPV4 packets¶
This document provides benchmark test for NIC VFs which are created from
kernel PFs or. These tests use l3fwd as a simple forwarder between NIC vfs.
The goal of this test plan is to have a tested benchmark between NIC vfs.
The Layer-3 Forwarding performance results are produced using l3fwd application.
For detail test plan, please refer to l3fwd_test_plan.rst.
193.1. Prerequisites¶
193.1.1. Topology¶
- It requires at least 1 port connected traffic generator::
- Port0 — TG0
- 2 ports::
- Port0 — TG0 Port1 — TG1
- 4 ports::
- Port0 — TG0 Port1 — TG1 Port2 — TG2 Port3 — TG3
193.1.2. Hardware¶
This suite focus on l3fwd application, so any standard Ethernet Network Adapter is qualified.
193.1.3. Software¶
dpdk: git clone http://dpdk.org/git/dpdk trex: git clone http://trex-tgn.cisco.com/trex/release/v2.93.tar.gz
193.2. Test Case¶
The test case check the throughput result with lpm mode and ipv4, in the case, we will send the bi-direction flows with line rate, then we can check the passthrough rate.
The l3fwd application has a number of command line options, here list the key options will be tested:
./dpdk-l3fwd [EAL options] -- -p PORTMASK
[--force-max-simd-bitwidth=max-simd-bitwidth]
--config(port,queue,lcore)[,(port,queue,lcore)]
[--rx-queue-size NPKTS]
[--tx-queue-size NPKTS]
[--parse-ptype]
[-L]|[-E]
...
Note:
--force-max-simd-bitwidth: This setting is used to determine the vector path for component selection.
And the default is avx2.
--rx-queue-size: Rx queue size in decimal and default is 1024.
--tx-queue-size: Tx queue size in decimal and default is 1024.
--parse-ptype: Optional, set to use software to analyze packet type.
Without this option, hardware will check the packet type.
[-L]|[-E]: L3fwd uses the parameters "-L" and "-E" to correspond to LPM and EM modes respectively.
And the default is LPM mode.
193.2.1. Common Steps¶
Bind PF ports to kernel driver, i40e or ice, then create 1 VF from each PF, take E810 for example:
<dpdk_dir>#echo 1 > /sys/bus/pci/devices/0000\:17\:00.0/sriov_numvfs
Set vf mac address:
<dpdk_dir>#ip link set ens5f0 vf 0 mac 00:12:34:56:78:01
Bind all the created VFs to vfio-pci:
<dpdk_dir>#./usertools/dpdk-devbind.py -s 0000:17:00.0 'Device 1592' if=ens5f0 drv=ice unused=vfio-pci 0000:17:01.0 'Device 1592' if=ens5f0v0 drv=iavf unused=vfio-pci <dpdk_dir>#./usertools/dpdk-devbind.py -b vfio-pci <pci device id> <dpdk_dir>#./usertools/dpdk-devbind.py -b vfio-pci 0000:17:01.0
Configure traffic generator to send traffic
- Routing table for IPv4 packets
- In LPM mode, the LPM table used for packet routing is:
# LPM prefix (IP/length) 0 198.18.0.0/24 1 198.18.1.0/24 2 198.18.2.0/24 3 198.18.3.0/24 - The flows need to be configured and started by the traffic generator:
- IPv4 packets
Flow Traffic Gen. Port IPv4 Dst. Address IPv4 Src. Address 1 TG0 198.18.1.1 Any Ip 2 TG1 198.18.0.1 Any Ip 3 TG2 198.18.3.1 Any Ip 4 TG3 198.18.2.1 Any Ip Set the packet length : 64 bytes-1518 bytes The IPV4 Src Address increase with the num 1024.
Test result table
Fwd_core Frame Size Throughput Rate 1C/1T 64 xxxxx Mpps xxx % 1C/1T ... xxxxx Mpps xxx % 2C/2T 64 xxxxx Mpps xxx % 2C/2T ... xxxxx Mpps xxx % 4C/4T 64 xxxxx Mpps xxx % 4C/4T ... xxxxx Mpps xxx % 8C/8T 64 xxxxx Mpps xxx % 8C/8T ... xxxxx Mpps xxx %
193.2.2. Test Case: test_perf_vf_throughput_ipv4_lpm¶
Bind VF ports to dpdk driver as common step 1.
Launch l3fwd:
<build_dir>/examples/dpdk-l3fwd -l 5-8 -n 8 --force-max-simd-bitwidth=512 \ -- -p 0x1 --config "(0,0,5),(0,1,6),(0,2,7),(0,3,8)" --rx-queue-size 1024 \ --tx-queue-size 1024 --parse-ptype
Configure traffic generator to send traffic as common step 2.
Record Test results as common step 3.