Episode 25 – MPLS Part 1

In a new protocol deep-dive series, Nick Russo and Russ White return to discuss MPLS. In part one, we discuss the primary use cases for MPLS, label allocation, and what SD-WAN means for the future of MPLS.

Show Notes:

1. MPLS solves 3 fundamental problems, individually or in concert
   1. Multi-tenancy/VPNs
   2. Traffic engineering
   3. Fast reroute
2. 4 bytes in a shim header, technically not a label, but we call it that
   1. 20 bits for label value, 2^20 ~= 1 million values (this is important)
   2. 3 bits for EXP, QoS really
   3. 1 S-bit to signal bottom of stack
   4. 8 bits TTL
3. Label depth is theoretically infinite, but some HW platforms have a tolerance
4. Many ways to allocate labels
   1. LDP transport
   2. LDP pseudowire
   3. BGP labeled unicast
   4. BGP based IP VPNs (VPNv4/v6)
   5. BGP pseudowire
   6. SR (really built into OSPF and ISIS for distribution)
   7. RSVP-TE
5. Some forward rules are worth mentioning (basic LDP/BGP-LU environment)
   1. If route learned via IGP/static, LDP label must be used
   2. If route learned via BGP, BGP label must be used
   3. No exceptions
6. Penultimate Hop Popping: second to last hop removes topmost label when signaled with imp-null from last hop along a given LSP, saves a lookup
7. Is MPLS is a tunnel or not:
   1. Nick says always
   2. Russ says sometimes, depending on label depth
8. Dispel rumor: MPLS is a technology, not a service. It’s incorrect to ask “Will SD-WAN supplant MPLS?” This is akin to saying “Will pizza delivery service supplant water?” A more reasonable question would be “Will SD-WAN supplant private WANs?”