Problem Description

Follow Topology

 10.200.0.24               10.200.0.26    10.200.0.32
   +------+      +------+     +------+     +------+
   |XiAn  |      | SZ   |     |HKBR  |     |Canada|
   |Huawei+------+      +-----+ASR9k +-----+ASR9k |
   +------+      +------+     +------+     +------+

For the issue, HuaWei is Head, asr9k should mid and end node. And Huawei found they put the traffics to Canada by LDP label, not TE label. So traffics should arrive to HK ASR9k by LDP label, then forwarding by LDP too from HK ASR9k to Canada. That not match normal scenario.

For normal scenario, should use TE label from head to end. And after Huawei put traffics to TE LSP, the issue will recovery to normal. Refer to why Huawei send traffics by LDP, that should their issue, will fix in future by them, but even if to do that, 9k shouldn’t drop packets. We need find whether packets drop at 9k and due to label issue first. Now set up test environment in customer site. waiting update.

完整阅读

Introduction

Customer run PBB-EVPN on 9001 NV at different site, and interconnect their different DC. Due to only have 1 interconnect link between 9001 NV, so they need a standby link that though Internet by GRE. After checked in lab, EVPN over GRE looks like same as L2VPN over GRE (support from 4.3 on ASR9k), and I had completed test for EVPN over GRE in lab. I will share config and topology.

But please attention: After checked, PBB-EVPN over GRE not offical release, so not suggest do it for customer. For the articles, not only talk about EVPN over GRE, and include how to troubleshooting PBB-EVPN

做为工程师，常常遇到一些协议交互的问题，需要确认数据包的具体信息，这时常常会用到几种方法：
1. SPAN抓包
对于这种方法，结果分析起来最方便，但操作过程最麻烦
2. debug
这种方法最直观，但是debug数据如果非常多，会影响设备的正常运行

下面就是采用其他方法来达成这种需求，虽然用的是udp来说明，但是同样适用于TCP和RAW：

RP/0/RP1/CPU0:CRS2(config)#udp directory /tmp/udp
RP/0/RP1/CPU0:CRS2(config)#commit
RP/0/RP1/CPU0:CRS2(config)#ipv4 access-list hsrp-packet
RP/0/RP1/CPU0:CRS2(config-ipv4-acl)#20 permit udp any eq 1985 any eq 1985
RP/0/RP1/CPU0:CRS2(config-ipv4-acl)#30 deny ipv4 any any
RP/0/RP1/CPU0:CRS2(config-ipv4-acl)#exit
RP/0/RP1/CPU0:CRS2(config)#ipv6 access-list v6-filter
RP/0/RP1/CPU0:CRS2(config-ipv6-acl)#10 deny ipv6 any any
RP/0/RP1/CPU0:CRS2(config-ipv6-acl)#exit
RP/0/RP1/CPU0:CRS2(config)#commit
RP/0/RP1/CPU0:CRS2(config)#exit
RP/0/RP1/CPU0:CRS2#debug udp packet v4-access-list hsrp-packet v6-access-list v6-filter hex control-block location x/x/cpu0

You can check the capture by follow patch:
RP/0/RP1/CPU0:CRS2#run
# cd /tmp/udp
#ls
#more xxxx

环境

 --------+--------------------+---------
         |   22.22.22.22/32   |
         |                    |
    +----+----+          +----+----+
    | 2.2.2.2 |          | 3.3.3.3 |
    | RouterA |          | RouterB |
    +-\----\--+          +-/---/---+
       \    \             /   /
        \\   \           /   /
          \   \         /  //
           \   \F2/0   /  /
            \\  \     /  /
         F1/0 \  \ F3/0 / F4/0
               *--\-/--*
               |1.1.1.1|
               | CoreA |
               +-------+

不限设备，所有运行IOS的设备，包括GSR，7609等。
在早期版本，不支持Multi hierarchical CEF，仅仅支持一层递归后的转发。这样产生了很多限制，例如今天提到的双PE结构。在特定版本后（包括IOS和IOX），CEF的行为有了改变，并且支持多层CEF。不过CEF的行为也要看平台，因为GSR上任何版本都不支持这种多层CEF。
完整阅读

When you found have output queue drop for CEF720 LC, you can check follow step:
1. which port have issue, whether at same ASIC.
2. check whether have other error or have qos on issue port.
3. whether hw queue is full.

Follow is TS example:

Problem description:

Output queue have drop
1. After checked by follow command:
– show tech
– show inter switching x/x
– show int x/x counter de
– show inter x/x summary
完整阅读