2026-05-14
RFC: RFC 3270
Published: 2002
Authors: F. Le Faucheur, L. Wu, B. Davie, S. Davari, P. Vaananen, R. Krishnan, P. Cheval, J. Heinanen
If you've ever bought an "MPLS circuit" from a telco, sent VoIP across a corporate WAN, or wondered why your packets seem to magically prioritize themselves between data centers, RFC 3270 is part of the reason. It's the document that taught MPLS how to honor the IP Differentiated Services (DiffServ) model — and it's a small masterpiece of pragmatic protocol design solving a problem that looks trivial until you try it.
The problem. DiffServ (RFC 2474/2475) marks IP packets with a 6-bit DSCP in the IPv4 ToS byte. Routers use that DSCP to pick a Per-Hop Behavior (PHB) — queue, drop precedence, scheduling weight. But MPLS routers don't look at the IP header. They forward on a 20-bit label and have only a 3-bit EXP field (since renamed "Traffic Class" in RFC 5462) to express QoS. Three bits, eight values — and DiffServ already defines more than eight PHBs in common use. How do you map a 6-bit world onto a 3-bit pipe without losing information that operators paid real money to preserve?
The two-LSP-flavor answer. RFC 3270 defines two ways to build a Label Switched Path, and they're conceptually elegant:
Tunneling models. The other half of the RFC, often more cited in operations, is how DSCP gets preserved across an MPLS cloud when packets are pushed, swapped, and popped. It defines three modes:
Why it still matters. Twenty-four years on, MPLS hasn't gone away — it's just hidden. Carrier Ethernet, mobile backhaul, MPLS-TP, and segment routing all inherit this DiffServ machinery. When your SD-WAN appliance honors DSCP across an underlay, or your hyperscaler's inter-region traffic gets prioritized treatment, the conceptual model of E-LSP vs L-LSP and Uniform vs Pipe is still the vocabulary engineers reach for. The penalty-of-mismarking debugging guide for any MPLS-backed VPN essentially is RFC 3270.
A small historical note. The EXP field was renamed to "Traffic Class" by RFC 5462 in 2009 precisely because of this RFC: once operators were using those bits for production QoS classification, calling them "experimental" was actively misleading vendors and customers. RFC 3270 didn't just standardize behavior; it forced an honest renaming of the header itself.
