Tuesday, 21 March 2017

Vx Lan Routing

Asymmetric Vs Symmetric IRB : 


  • EVPN integrated Routing (IRB) has two options: 
  • asymmetric IRB
  • symmetric IR

Asymmetric IRB 

  • Ingress VTEP does both L2 & L3 lookup
  • Egress VTEP does L2 lookup only
  • i.e Bridge - Route - Bridge 

Symmetric IRB

  • Ingress VTEP does both L2 & L3 lookup
  • Egress VTEP does both L3 & L2 Lookup
  • i.e Bridge-Route-Bridge

      A----(L2)--SVI--(L3)---SVI-------(L3)-------SVI----(L2)------B

Asymmetric IRB issues

VTEP must have all VNIs configured that require routing , result is increased ARP Cache and CAM
sizes , i.e Control plane scaling issue . 

VxLAN BGP EVPN With L3 VNIs : 

install feature-set virtualization 
install feature-set fabric
feature-set fabric
feature fabric forwarding 
feature nv overlay evpn
feature ospf 
feature bgp 
feature pim
feature interface-vlan
feature nv overlay
feature vn-segment-vlan-based
vlan 11
   vn-segment 11111
vlan 33
   vn-segment 33333
vrf context CUSTOMER1
   vni 33333
   rd auto
address-family ipv4 unicast
   route-target both auto
   route-target both auto evpn

interface vlan 11 
no shut
vrf member 11.0.0.0.254/24
fabric forwarding mode anycast-gateway
!
interface vlan 33 
no shut
vrf member CUSTOMER1
ip forward 
!
interface nve1 
no shut
source-interface loopback0
host-reachability protocol bgp 
member vni 11111
mcast-group 224.11.11.11
member vni 33333 associate-vrf 
route-map PERMIT permit 10 
!
router bgp 12345
    neighbor 1.1.1.71
    remote-as 12345
update-source loopback0
address-family l2vpn evpn
send-community both 
!
neighbor 1.1.1.72 
remote-as 12345
update-source loopback0
address-family l2vpn evpn 
send-community both
!
vrf CUSTOMER1

evpn 
  vni 11111 l2 
   rd auto
  route-target import auto
  route-target export auto
vrf defalt 


                                      vPC & VxLAN

vPC & VxLAN BGP traffic Flow problems : 
  • Vxlan traffic is tunneled over the overlay network using the BGP next-hop address of the remote VTEP
  • NVE source interface (i.e loopback 0 ) is the default BGP next-hop for advertised routes
  • In a vPC , both vPC peers advertise duplicate EVPN MAC / IP routes to spine RRs.
  • With other attributes equal , next-hop is tie breaker in BGP best path selection 
  • Implies that one vPC peer is always preferred for dual attached hosts.
  • Result is the egress traffic from vPC Member is load balanced , but ingress traffic is polarized
  • Workaround is to use Anycast VTEP address 

                                        vPC Anycast VTEP

vPC peers share duplicate ip address on NVE source interface 
  • Peer1 - interface loopback0 ; ip address 1.1.1.51/32
  • Peer2 - interface loopback0 ; ip address 1.1.1.52/32
  • Both peers - interface loopback0 ; ip address 1.1.1.111/32 secondary
BGP Next-hop is automatically set to secondary address for locally originated routes .
  • i.e L2VPN EVPN MAC/IP routes for vPC Member ports

                         Nexus 5600 & NVE Peer-Link-vlan 

  • On nexus 5600 , all traffic across the vPC Peer Link must be Vxlan encapsulated due to ASIC implementation 
  • Normal vPC Peer Link is a classical ethernet trunk 
  1. Result is that East/West flows over vPC Peer Link all broken by default 
  2. i.e , the VNI number is lost when pkt is sent out peer link 
  • Peer Link is normally only used for orphans or in failure scenarios 
  • Result is that everything looks fine until the failure occures 
  • Traffic to orphans & single attached members black holed over vPC Peer Link
  • Workaround is to maintain VxLan encapsulation across peer link
  • implemented as "vpc nve peer-link-vlan"

                Configuring NVE Peer-Link-Vlan

  • Create new VLAN & Specify as NVE Peer Link VLAN ( vlan 999; vpc nve peer-link-vlan 999)
  • Establish layer 3 peering across NVE peer link VLAN (interface vlan 999 ; ip router ospf 1 area 0 )
  • Traffic engineering so other vPC Peer's VTEP loopback is preferred over vPC Peer link 
  1. ip ospf cost 10
  2. isis metric 10 level-2


--------------------------------------ooooooooooooooo----------------------------------------------------
































No comments:

Post a Comment