Understanding MPLS: Principles, Architecture, and Implementation

Introduction

MPLS (Multiprotocol Label Switching) emerged in the mid‑1990s to address the performance limitations of IP routing, combining the speed of ATM’s fixed‑length labels with the flexibility of IP networks. It was originally designed to accelerate router forwarding by using short, locally significant labels instead of full IP header lookups at every hop.

Origin of MPLS

Rapid growth of Internet traffic highlighted the bottleneck of software‑based longest‑match routing. ATM offered high‑speed label switching but was costly and complex, prompting the development of MPLS to merge IP’s simplicity with ATM’s performance.

Definition

MPLS sits between the link layer and network layer, providing a connection‑oriented service to IP while receiving services from the link layer. It replaces IP forwarding with label switching, using short, locally significant identifiers similar to ATM VPI/VCI or Frame Relay DLCI, and supports multiple underlying link‑layer technologies.

How MPLS Works

The operation of MPLS consists of two parts: the internal architecture of a single device and the network‑wide structure of multiple devices.

Architecture

MPLS architecture is divided into a control plane and a forwarding (data) plane:

The control plane is connection‑less and handles label allocation, LFIB construction, LSP establishment, and teardown.

The forwarding plane is connection‑oriented, using ATM, Ethernet, etc., to add or remove labels on IP packets and forward them based on the label forwarding information base.

The typical device diagram (Figure 1‑1) shows the interaction between IP routing, label distribution, and forwarding.

Network Structure

A typical MPLS network comprises Label Switching Routers (LSRs). Edge LSRs are called Label Edge Routers (LERs), while interior LSRs are Core LSRs.

Implementation Principles

MPLS LSP

An LSP (Label Switched Path) is a unidirectional path that packets follow. It has three types of nodes:

Ingress – the entry point that pushes the first label onto the packet.

Transit – intermediate nodes that swap labels according to their LFIB.

Egress – the exit point that pops the label and restores the original packet.

Ingress and egress nodes act as both LSRs and LERs; transit nodes are pure LSRs.

Labels

A label is a 4‑byte header consisting of four fields:

Label (20 bits) – the actual label value.

Exp (3 bits) – used for Class of Service.

BoS (1 bit) – Bottom of Stack indicator.

TTL (8 bits) – Time‑to‑Live, same as in IP.

Labels are placed between the link‑layer header and the IP header, enabling any link‑layer technology to carry them. Multiple labels can be stacked (LIFO order), allowing nested encapsulation.

Label Operations

Push – insert a new label at the top of the stack.

Swap – replace the top label with a new one.

Pop – remove the top label, often performed at the penultimate hop (PHP) to reduce load on the egress router.

Establishing LSPs

Labels are allocated downstream and propagated upstream to build LFIB entries, forming an LSP. LSPs can be static (manually configured) or dynamic (established via routing and label distribution protocols such as LDP, RSVP‑TE, or MP‑BGP). LDP is the most widely used label distribution protocol.

References

Source: Huawei MPLS Documentation