环境介绍
一个简单的CE双上连的VPN环境，客户路由8.8.8.8/32 将送到r1。
r1 配置maximum-path，使去往ce的流量可以在r2/r3 之间负载。
r2/r3 我们都配置了eibgp maximum-paths，那么r2/r3 都会存在两条路由负载（一条是从ibgp学了vpn路由， 一条是从vrf下的ebgp学来的路由）。
RP/0/0/CPU0:R2-AS56704#show route vrf 1
B 8.8.8.8/32 [20/0] via 10.1.3.3 (nexthop in vrf default), 00:21:10
[20/0] via 10.1.24.4, 00:21:10
label mode per-prefix （default）
默认情况r2/r3 都是default的标签分配方式， 即不同的vpn路由分配的标签都不一样。
RP/0/0/CPU0:R1-AS56704#show cef vrf 1 8.8.8.8/32 | in label
Sun Nov 5 16:07:25.162 UTC
next hop 10.1.12.2/32 Gi0/0/0/0 labels imposed {ImplNull 24003}
next hop 10.1.13.3/32 Gi0/0/0/1 labels imposed {ImplNull 24003}
以去往r2的流量为例，其vpn标签为24003。数据包到达r2 后会查询标签转发表，如下:
RP/0/0/CPU0:R2-AS56704#show mpls forwarding labels 24003
Sun Nov 5 16:08:40.506 UTC
Local Outgoing Prefix Outgoing Next Hop Bytes
Label Label or ID Interface Switched
------ ----------- ------------------ ------------ --------------- ------------
24003 Unlabelled 8.8.8.8/32[V] Gi0/0/0/1 10.1.24.4 0
24003 8.8.8.8/32[V] 10.1.3.3 0 (!)
RP/0/0/CPU0:R2-AS56704#show mpls forwarding labels 24003 detail
Sun Nov 5 16:18:55.223 UTC
Local Outgoing Prefix Outgoing Next Hop Bytes
Label Label or ID Interface Switched
------ ----------- ------------------ ------------ --------------- ------------
24003 Unlabelled 8.8.8.8/32[V] Gi0/0/0/1 10.1.24.4 0
Updated: Nov 5 16:18:48.293
Path Flags: 0x60a0 [ EXT MULT ]
Version: 38, Priority: 3
Label Stack (Top -> Bottom): { Unlabelled }
NHID: 0x0, Encap-ID: N/A, Path idx: 1, Backup path idx: 0, Weight: 0
MAC/Encaps: 14/14, MTU: 1500
Outgoing Interface: GigabitEthernet0/0/0/1 (ifhandle 0x00000060)
Packets Switched: 0
24004 8.8.8.8/32[V] 10.1.3.3 0 (!)
Updated: Nov 5 16:18:48.293
Path Flags: 0x6180 [ BKUP, NoFwd MULT ] <<<<<<<<
Version: 38, Priority: 3
Label Stack (Top -> Bottom): { 24004 }
NHID: 0x0, Encap-ID: N/A, Path idx: 0, Backup path idx: 0, Weight: 0
MAC/Encaps: 0/4, MTU: 0
Packets Switched: 0
(!): FRR pure backup <<<<<<<<
（!）为备份路径，流量只会走第一条，剥离标签，直接从出接口Gi0/0/0/1 送出。
同理r3上的标签转发表也类似也是剥离标签从出接口Gi0/0/0/0 送出。
RP/0/0/CPU0:R3-AS56704#show mpls forwarding labels 24003
Sun Nov 5 16:11:25.436 UTC
Local Outgoing Prefix Outgoing Next Hop Bytes
Label Label or ID Interface Switched
------ ----------- ------------------ ------------ --------------- ------------
24003 Unlabelled 8.8.8.8/32[V] Gi0/0/0/0 10.1.34.4 0
24003 8.8.8.8/32[V] 10.1.2.2 0 (!)
label mode per-vrf
RP/0/0/CPU0:R2-AS56704#show configuration commit changes last 1
Mon Nov 6 13:47:02.326 UTC
Building configuration...
!! IOS XR Configuration 6.3.1
router bgp 56704
vrf 1
address-family ipv4 unicast
label mode per-vrf
将r2/r3 的标签分配方式 改为per-vrf， 这个时候我们看下r1 上显示的标签如下:
RP/0/0/CPU0:R1-AS56704#show cef vrf 1 8.8.8.8/32 | in label
Sun Nov 5 16:28:56.814 UTC
next hop 10.1.12.2/32 Gi0/0/0/0 labels imposed {ImplNull 24004}
next hop 10.1.13.3/32 Gi0/0/0/1 labels imposed {ImplNull 24004}
流量如果到了r2 上， 会先查mpls 标签表24004，如下； 它会剥离标签， 最后再查找下cef表，看看究竟该从哪个接口送出。
RP/0/0/CPU0:R2-AS56704#show mpls forwarding labels 24004
Sun Nov 5 16:29:43.769 UTC
Local Outgoing Prefix Outgoing Next Hop Bytes
Label Label or ID Interface Switched
------ ----------- ------------------ ------------ --------------- ------------
24004 Aggregate 1: Per-VRF Aggr[V] 1 0
当你查看cef表就会发现， 由于eibgp maximum-paths的配置， 我们将有两个路由负载，也就是说部分流量不会直接送往ce，将会带标签送到r3.
RP/0/0/CPU0:R2-AS56704#show route vrf 1
B 8.8.8.8/32 [20/0] via 10.1.3.3 (nexthop in vrf default), 00:00:43
[20/0] via 10.1.24.4, 00:00:43
RP/0/0/CPU0:R2-AS56704#show cef vrf 1 8.8.8.8/32 detail
Sun Nov 5 16:31:23.022 UTC
8.8.8.8/32, version 42, internal 0x1000001 0x0 (ptr 0xa11dd798) [1], 0x0 (0x0), 0x208 (0xa197d0e8)
Updated Nov 5 16:19:32.900
Prefix Len 32, traffic index 0, precedence n/a, priority 3
gateway array (0xa111688c) reference count 1, flags 0x102878, source rib (7), 0 backups
[1 type 1 flags 0x48441 (0xa15ed298) ext 0x0 (0x0)]
LW-LDI[type=0, refc=0, ptr=0x0, sh-ldi=0x0]
gateway array update type-time 1 Nov 5 15:31:29.808
LDI Update time Nov 5 15:31:29.808
via 10.1.3.3/32, 3 dependencies, recursive, bgp-multipath, backup [flags 0x6180]
path-idx 0 NHID 0x0 [0xa1645798 0x0]
recursion-via-/32
next hop VRF - 'default', table - 0xe0000000
next hop 10.1.3.3/32 via 24001/0/21
next hop 10.1.23.3/32 Gi0/0/0/2 labels imposed {ImplNull 24004}
via 10.1.24.4/32, 4 dependencies, recursive, bgp-ext, bgp-multipath [flags 0x60a0]
path-idx 1 NHID 0x0 [0xa11dd824 0x0]
next hop 10.1.24.4/32 via 10.1.24.4/32
next hop 10.1.24.4/32 Gi0/0/0/1 labels imposed {None}
Load distribution: 0 1 (refcount 1)
Hash OK Interface Address
0 Y Unknown 24001/0
1 Y recursive 10.1.24.4
而如果r3也都配置了eigbp maximum-paths和lable-mode per-vrf， r3上也会先查标签表剥离标签，查cef表有两条路径负载， 环路产生（r2 和 r3之间）。
RP/0/0/CPU0:R3-AS56704#show mpls forwarding labels 24004
Sun Nov 5 16:34:00.653 UTC
Local Outgoing Prefix Outgoing Next Hop Bytes
Label Label or ID Interface Switched
------ ----------- ------------------ ------------ --------------- ------------
24004 Aggregate 1: Per-VRF Aggr[V] 1 0
RP/0/0/CPU0:R3-AS56704#show cef vrf 1 8.8.8.8/32 detail
Sun Nov 5 16:36:28.863 UTC
8.8.8.8/32, version 35, internal 0x1000001 0x0 (ptr 0xa11e55f4) [1], 0x0 (0x0), 0x208 (0xa197d0b8)
Updated Nov 5 16:19:33.553
Prefix Len 32, traffic index 0, precedence n/a, priority 3
gateway array (0xa111e6e8) reference count 1, flags 0x102878, source rib (7), 0 backups
[1 type 1 flags 0x48441 (0xa15ed298) ext 0x0 (0x0)]
LW-LDI[type=0, refc=0, ptr=0x0, sh-ldi=0x0]
gateway array update type-time 1 Nov 5 15:30:39.934
LDI Update time Nov 5 15:30:39.934
via 10.1.2.2/32, 3 dependencies, recursive, bgp-multipath, backup [flags 0x6180]
path-idx 0 NHID 0x0 [0xa164570c 0x0]
recursion-via-/32
next hop VRF - 'default', table - 0xe0000000
next hop 10.1.2.2/32 via 24000/0/21
next hop 10.1.23.2/32 Gi0/0/0/2 labels imposed {ImplNull 24004}
via 10.1.34.4/32, 4 dependencies, recursive, bgp-ext, bgp-multipath [flags 0x60a0]
path-idx 1 NHID 0x0 [0xa11e5680 0x0]
next hop 10.1.34.4/32 via 10.1.34.4/32
next hop 10.1.34.4/32 Gi0/0/0/0 labels imposed {None}
Load distribution: 0 1 (refcount 1)
Hash OK Interface Address
0 Y Unknown 24000/0
1 Y recursive 10.1.34.4
而如果你只在r2上修改label 分配方式为per-vrf， r3 上保持默认的标签分配方式per-prefix。
那即使流量在r2上仍然会有一半送到r3， 但由于per-prefix的标签分配方式直接有出接口， 在r3 上会直接转发给ce，这样也不会出环。
label mode per-ce
per-ce的标签分配方式，其实和per-prefix类似， 它是有直接的出接口的， 流量也不需要经过二次查找， 所以没成环的风险。
RP/0/0/CPU0:R2-AS56704#show mpls forwarding
Sun Nov 5 16:43:33.192 UTC
Local Outgoing Prefix Outgoing Next Hop Bytes
Label Label or ID Interface Switched
------ ----------- ------------------ ------------ --------------- ------------
24006 Unlabelled No ID Gi0/0/0/1 10.1.24.4 0
Aggregate No ID 1 0 (!)
24006 Unlabelled No ID Gi0/0/0/1 10.1.24.4 0
Updated: Nov 5 16:39:36.928
Path Flags: 0x6080 [ MULT ]
Version: 15, Priority: 3
Label Stack (Top -> Bottom): { Unlabelled }
NHID: 0x0, Encap-ID: N/A, Path idx: 1, Backup path idx: 0, Weight: 0
MAC/Encaps: 14/14, MTU: 1500
Outgoing Interface: GigabitEthernet0/0/0/1 (ifhandle 0x00000060)
Packets Switched: 0
Aggregate No ID 1 0 (!)
Updated: Nov 5 16:39:36.928
Path Flags: 0x100 [ BKUP, NoFwd ]
Label Stack (Top -> Bottom): { }
MAC/Encaps: 0/0, MTU: 0
Packets Switched: 0
标签分配方式的总结
- per-prefix 默认的标签分配方式， 缺点为如果分配的标签量很大的情况下，标签的收敛上相对于ip的收敛会慢上很多。
- per-vrf 需要二次查找路由表
eibgp的使用场景
粗略画一个图，siteA有两个CE分别连到一个PE；siteB有两个CE也分别连到一个PE。
对于siteA-CE1而言， 要访问siteB的业务， 默认只会走PE1→ siteB-CE1, 为了充分的利用PE2和siteB-CE2之间的链路我们可以在PE1上开启eibgp maximum-paths；
同理PE2上也可以开启eibgp maximum-paths。
No comments
Comments feed for this article