Skip to main content

RFC 4379 - 2. Motivation

2. Motivation​

When an LSP fails to deliver user traffic, the failure cannot always be detected by the MPLS control plane. There is a need to provide a tool that would enable users to detect such traffic "black holes" or misrouting within a reasonable period of time, and a mechanism to isolate faults.

In this document, we describe a mechanism that accomplishes these goals. This mechanism is modeled after the ping/traceroute paradigm: ping (ICMP echo request [ICMP]) is used for connectivity checks, and traceroute is used for hop-by-hop fault localization as well as path tracing. This document specifies a "ping" mode and a "traceroute" mode for testing MPLS LSPs.

The basic idea is to verify that packets that belong to a particular Forwarding Equivalence Class (FEC) actually end their MPLS path on a Label Switching Router (LSR) that is an egress for that FEC. This document proposes that this test be carried out by sending a packet (called an "MPLS echo request") along the same data path as other packets belonging to this FEC. An MPLS echo request also carries information about the FEC whose MPLS path is being verified. This echo request is forwarded just like any other packet belonging to that FEC. In "ping" mode (basic connectivity check), the packet should reach the end of the path, at which point it is sent to the control plane of the egress LSR, which then verifies whether it is indeed an egress for the FEC. In "traceroute" mode (fault isolation), the packet is sent to the control plane of each transit LSR, which performs various checks that it is indeed a transit LSR for this path; this LSR also returns further information that helps check the control plane against the data plane, i.e., that forwarding matches what the routing protocols determined as the path.

One way these tools can be used is to periodically ping an FEC to ensure connectivity. If the ping fails, one can then initiate a traceroute to determine where the fault lies. One can also periodically traceroute FECs to verify that forwarding matches the control plane; however, this places a greater burden on transit LSRs and thus should be used with caution.

2.1. Use of Address Range 127/8​

As described above, LSP ping is intended as a diagnostic tool. It is intended to enable providers of an MPLS-based service to isolate network faults. In particular, LSP ping needs to diagnose situations

where the control and data planes are out of sync. It performs this by routing an MPLS echo request packet based solely on its label stack. That is, the IP destination address is never used in a forwarding decision. In fact, the sender of an MPLS echo request packet may not know, a priori, the address of the router at the end of the LSP.

Providers of MPLS-based services also need the ability to trace all of the possible paths that an LSP may take. Since most MPLS services are based on IP unicast forwarding, these paths are subject to equal-cost multi-path (ECMP) load sharing.

This leads to the following requirements:

  1. Although the LSP in question may be broken in unknown ways, the likelihood of a diagnostic packet being delivered to a user of an MPLS service MUST be held to an absolute minimum.

  2. If an LSP is broken in such a way that it prematurely terminates, the diagnostic packet MUST NOT be IP forwarded.

  3. A means of varying the diagnostic packets such that they exercise all ECMP paths is thus REQUIRED.

Clearly, using general unicast addresses satisfies neither of the first two requirements. A number of other options for addresses were considered, including a portion of the private address space (as determined by the network operator) and the newly designated IPv4 link local addresses. Use of the private address space was deemed ineffective since the leading MPLS-based service is an IPv4 Virtual Private Network (VPN). VPNs often use private addresses.

The IPv4 link local addresses are more attractive in that the scope over which they can be forwarded is limited. However, if one were to use an address from this range, it would still be possible for the first recipient of a diagnostic packet that "escaped" from a broken LSP to have that address assigned to the interface on which it arrived and thus could mistakenly receive such a packet. Furthermore, the IPv4 link local address range has only recently been allocated. Many deployed routers would forward a packet with an address from that range toward the default route.

The 127/8 range for IPv4 and that same range embedded in as IPv4- mapped IPv6 addresses for IPv6 was chosen for a number of reasons.

RFC 1122 allocates the 127/8 as "Internal host loopback address" and states: "Addresses of this form MUST NOT appear outside a host." Thus, the default behavior of hosts is to discard such packets. This

helps to ensure that if a diagnostic packet is misdirected to a host, it will be silently discarded.

RFC 1812 [RFC1812] states:

A router SHOULD NOT forward, except over a loopback interface, any packet that has a destination address on network 127. A router MAY have a switch that allows the network manager to disable these checks. If such a switch is provided, it MUST default to performing the checks.

This helps to ensure that diagnostic packets are never IP forwarded.

The 127/8 address range provides 16M addresses allowing wide flexibility in varying addresses to exercise ECMP paths. Finally, as an implementation optimization, the 127/8 provides an easy means of identifying possible LSP packets.