# Kireeti Kompella Bhupesh Kothari Thomas Spencer

## Presentation on theme: "Kireeti Kompella Bhupesh Kothari Thomas Spencer"— Presentation transcript:

Kireeti Kompella Bhupesh Kothari Thomas Spencer
BGP VPLS Multi-homing Kireeti Kompella Bhupesh Kothari Thomas Spencer

Problem Statement CE2 PE3 PE1 BGP VPLS CE1 PE2 PE4 CE3
CE1 wants resilient connectivity as a VPLS CE, so it is dual-homed to PE1 and PE2 Simple dual connectivity leads to loops and duplicate packets. One answer is to run STP on CE1

Solution Let PE1 and PE2 know that they are connected to the same site
Configure the VPLS VE ID to be the same Use BGP path selection to arbitrate among PE1 and PE2 as to which should be used to reach CE1 And which forwards packets from CE1

Multi-homing in 4364 VPNs CE2 LP=200 PE3 PE1 BGP VPN 10.1/16 CE1
CE1 wants resilient connectivity as a VPN CE, so it is dual-homed to PE1 and PE2 Choices: pick one of PE1 or PE2 to service CE1; or use both (ECMP)

Differences in VPLS Cannot do ECMP: must pick one of PE1 or PE2 (say PE1), and must do so consistently from all “other” PEs Both PE3 and PE4 must both choose PE1 In particular, the “losing” PE (here, PE2) must know that it lost The “winning” PE (here, PE1) is called the ‘designated PE’

Changes to BGP Path Selection
When are prefixes comparable? What part of the NLRI is to be compared? Who wins, and how can we ensure that everyone picks consistently? Simple answer: don’t use IGP metric in the tie-breaking rules Control? (via Local Preference) Details in draft

Result Designated PE forwards packets from and to CE
Non-designated PEs (losing PEs) drop packets from CE as well as from other PEs The effect is as if CE was single-homed to just the designated PE

Re-homing If the designated PE fails, or if the CE-PE link breaks, the designated PE withdraws its “route” All other PEs (including non-designated PEs) redo path selection and pick a new designated PE Traffic is re-established via the new PE

Effect of Route Reflection
CE2 PE3 PE1 RR CE1 PE2 PE4 CE3 If there is an RR in the path, the RR also does path selection; this can affect convergence. To get around this, use different RDs on each PE (as in 4364 VPNs)

Status RFC 4761, section 3.5 talks about multi-homing and how to accomplish it However, there aren’t enough details on how prefixes are to be compared, and how BGP path selection should be done This draft attempts to rectify this

Questions?