网络工程 - Trex 收不到任何数据包 - 吾爱随笔录

这是我对服务器图进行基准测试的设置

Trex 版本 v2.88

Trex 服务器 MAC

ens2f0: f8:f2:1e:5c:bf:78
ens2f1: f8:f2:1e:5c:bf:79

DUT 服务器 MAC

eth0: fa:16:3e:1c:be:82
eth1: fa:16:3e:d5:37:b1

trex_cfg.yaml

cat /etc/trex_cfg.yaml
Config file generated by dpdk_setup_ports.py
- version: 2
  #interfaces: ['07:00.0', '07:00.1']
  interfaces: ['ens2f0', 'ens2f1']
  port_info:
      - ip: 10.66.0.100
        default_gw: 10.66.1.33
      - ip: 10.68.0.100
        default_gw: 10.68.3.217

  platform:
      master_thread_id: 0
      latency_thread_id: 8
      dual_if:
        - socket: 0
          threads: [1,2,3,4,5,6,7,16,17,18,19,20,21,22,23]

我的场景文件

- duration : 5
  generator :
          distribution : "seq"
          clients_start : "16.0.0.1"
          clients_end   : "16.0.0.255"
          servers_start : "48.0.0.1"
          servers_end   : "48.0.255.255"
          clients_per_gb : 10
          min_clients    : 10
          dual_port_mask : "1.0.0.0"
          tcp_aging      : 0
          udp_aging      : 0
  cap_info :
     - name: cap2/rtp_160k.pcap
       cps : 10.0
       ipg : 10000
       rtt : 10000
       w   : 4
     - name: cap2/rtp_250k_rtp_only_1.pcap
       cps : 10.0
       ipg : 10000
       rtt : 10000
       w   : 4
     - name: cap2/rtp_250k_rtp_only_2.pcap
       cps : 10.0
       ipg : 10000
       rtt : 10000
       w   : 4

运行命令

./t-rex-64 --cfg /etc/trex_cfg.yaml -f /tmp/sfr.yaml -d 10 -m 1 -c 1 -k 2 --nc

在测试中，我看到 Trex 发送 UDP 流，但 RX 端没有收到任何数据，结果是 100% 丢包。当我在 DUT 设备上运行 tcpdump 时，我可以看到它使用正确的 MAC 地址将 UDP 流发送到 Trex RX 端。fa:16:3e:d5:37:b1 > f8:f2:1e:5c:bf:79

DUT 服务器 tcpdump

   16.0.0.17.58005 > 48.0.0.17.1024: [no cksum] UDP, length 1231
15:38:34.237250 fa:16:3e:d5:37:b1 > f8:f2:1e:5c:bf:79, ethertype IPv4 (0x0800), length 378: (tos 0x0, ttl 125, id 11822, offset 0, flags [none], proto UDP (17), length 364)
    16.0.0.18.9873 > 48.0.0.18.1026: [no cksum] UDP, length 336
15:38:34.237260 fa:16:3e:d5:37:b1 > f8:f2:1e:5c:bf:79, ethertype IPv4 (0x0800), length 1492: (tos 0x0, ttl 125, id 11852, offset 0, flags [none], proto UDP (17), length 1478)
    16.0.0.15.23195 > 48.0.0.15.1026: [no cksum] UDP, length 1450
15:38:34.237266 fa:16:3e:d5:37:b1 > f8:f2:1e:5c:bf:79, ethertype IPv4 (0x0800), length 434: (tos 0x0, ttl 125, id 11832, offset 0, flags [none], proto UDP (17), length 420)

测试一开始，我就看到了正确的 ARP 条目 mac-ip，但在 2 秒内它变得不完整，即使我的测试正在运行并继续发送数据。

[root@dut-2 ~]# arp -a | grep 10.68.0.100
? (10.68.0.100) at f8:f2:1e:5c:bf:79 [ether] on eth1
[root@dut-2 ~]# arp -a | grep 10.68.0.100
? (10.68.0.100) at <incomplete> on eth1
Any idea why ARP getting incomplete in 2 second?

更新：

当我在 DUT 服务器上运行 tcpdump 然后我发现 Trex RX 端接口请求 ARP 并且 DUI 回复但是当 DUT 请求请求然后 Trex 没有回复（10.68.3.217 是 DUT 服务器的 eth1 IP）

[root@dut-1 ~]# tcpdump -i eth1 -nn arp -e -vv
tcpdump: listening on eth1, link-type EN10MB (Ethernet), capture size 262144 bytes
16:03:26.527165 f8:f2:1e:5c:bf:79 > ff:ff:ff:ff:ff:ff, ethertype ARP (0x0806), length 60: Ethernet (len 6), IPv4 (len 4), Request who-has 10.68.0.100 (01:03:05:07:09:00) tell 10.68.0.100, length 46
16:03:26.527192 f8:f2:1e:5c:bf:79 > ff:ff:ff:ff:ff:ff, ethertype ARP (0x0806), length 60: Ethernet (len 6), IPv4 (len 4), Request who-has 10.68.3.217 (01:03:05:07:09:01) tell 10.68.0.100, length 46
16:03:26.527201 fa:16:3e:d5:37:b1 > f8:f2:1e:5c:bf:79, ethertype ARP (0x0806), length 42: Ethernet (len 6), IPv4 (len 4), Reply 10.68.3.217 is-at fa:16:3e:d5:37:b1, length 28
16:03:32.107249 fa:16:3e:d5:37:b1 > f8:f2:1e:5c:bf:79, ethertype ARP (0x0806), length 42: Ethernet (len 6), IPv4 (len 4), Request who-has 10.68.0.100 tell 10.68.3.217, length 28
16:03:33.131245 fa:16:3e:d5:37:b1 > f8:f2:1e:5c:bf:79, ethertype ARP (0x0806), length 42: Ethernet (len 6), IPv4 (len 4), Request who-has 10.68.0.100 tell 10.68.3.217, length 28
16:03:34.155245 fa:16:3e:d5:37:b1 > f8:f2:1e:5c:bf:79, ethertype ARP (0x0806), length 42: Ethernet (len 6), IPv4 (len 4), Request who-has 10.68.0.100 tell 10.68.3.217, length 28
16:03:35.183759 fa:16:3e:d5:37:b1 > ff:ff:ff:ff:ff:ff, ethertype ARP (0x0806), length 42: Ethernet (len 6), IPv4 (len 4), Request who-has 10.68.0.100 tell 10.68.3.217, length 28
16:03:36.203260 fa:16:3e:d5:37:b1 > ff:ff:ff:ff:ff:ff, ethertype ARP (0x0806), length 42: Ethernet (len 6), IPv4 (len 4), Request who-has 10.68.0.100 tell 10.68.3.217, length 28
16:03:37.227240 fa:16:3e:d5:37:b1 > ff:ff:ff:ff:ff:ff, ethertype ARP (0x0806), length

对于实验，如果我使用以下命令在服务器模式下运行 t-rex：./t-rex-64 -i --astf

然后我可以看到它创建命名空间，我可以 ping DUT ip 地址，还可以验证 16.0.0.0/8 和 48.0.0.0/8 路由，到目前为止一切看起来都很好。

[root@Trex v2.86]# ip netns list
trex-a-1-0 (id: 1)
trex-a-0-0 (id: 0)
[root@Trex v2.86]# ip netns exec trex-a-1-0 ip a
1: lo: <LOOPBACK> mtu 65536 qdisc noop state DOWN group default qlen 1000
    link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
43: trex-a-1-0-L@if44: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 9280 qdisc noqueue state UP group default qlen 1000
    link/ether f8:f2:1e:5c:bf:79 brd ff:ff:ff:ff:ff:ff link-netnsid 0
    inet 10.68.0.100/32 scope global trex-a-1-0-L
       valid_lft forever preferred_lft forever

但是当我使用常规命令运行时，它不会创建任何网络命名空间，这正常吗？

./t-rex-64 --cfg /etc/trex_cfg.yaml -f /tmp/spatel-sfr.yaml -d 60 -m 1 -c 1 -k 2 --nc