next up previous contents
Next: Multicast fault recovery Up: Resilience and protection in Previous: Network layer protection   Contents


MPLS Unicast Fast Reroute

Rerouting at the MAC and physical layer is fast but requires dedicated hardware. On the other hand, IP rerouting is slow but does not rely on any specific topology and is implemented in every router over the Internet. MPLS, which is implemented between the IP and MAC layers, supports rerouting mechanisms that provide a trade-off between repair speed and deployment cost.

Several methods have been proposed to reroute unicast traffic in MPLS [58]. We present here the fastest MPLS rerouting mechanism available, MPLS Fast Reroute [34]. A slower, less complex mechanism can be found in [64]. A comparison of different MPLS rerouting mechanisms can be found in [28].

Figure 2.5: Unicast Fast Reroute mechanism. When link (c) fails, the traffic that flows from the ingress LER to the egress LER is rerouted from the primary path to the backup path.
\includegraphics[width=\textwidth]{figures/fast_reroute}

Fast Reroute relies on pre-planning and requires that a backup path is computed and advertised before a link failure can be repaired. Figure 2.5 illustrates unicast Fast Reroute. Suppose traffic goes from the Ingress LER to the Egress LER of an MPLS domain through the primary LSP $(a, b, c, d, e)$, and that the backup LSP $(f, g, h, i)$ has already been set up. The first router of the backup path is called PSL (Path Switching LSR), and the last router of the backup path is called PML (Path Merging LSR). If link $c$ fails, the router $U$ upstream of $c$ detects the failure and sends the packets whose destination was the Egress router back to the Ingress router. When the first of those packets reaches the PSL, the PSL knows that a failure has occurred. Alternatively, $U$ could send a notification message to the PSL to let it know of the failure. The PSL then forwards on the backup path the packets coming back from $U$. This ensures that no packet is lost after the fault is detected by $U$, during the notification step of the rerouting mechanism. The switchover step is instantaneous as the PSL only needs to start forwarding the packets coming from the Ingress LSR going to the Egress LSR on the backup path instead of the primary path. A disadvantage of Fast Reroute is that the packets sent during the notification step arrive out of order. Also, some packets will cover up to three times the distance between the ingress and the egress router if the PSL is the ingress node, the PML is the egress node, and the failed link is the last link on the primary path before the PML. The major advantage of MPLS unicast Fast Reroute, however, is that rerouting is fast and no packet is lost after the fault is detected. When the failed link is physically repaired, node $U$ sends a notification message to the PSL which can send traffic back from the backup path to the primary path in the switchback step. Switchback, like switchover, is instantaneous.

MPLS Fast Reroute is faster than IP rerouting but slower than MAC or physical layer rerouting. Indeed, Fast Reroute saves the switchover step that is expensive in IP rerouting, but does not get rid of the notification step as lower layer mechanisms do. Detection can be performed as in SONET using dedicated hardware or like in routing by sending probes over the link regularly. Detection times are expected to be in the order of 1 to 100 milliseconds. Notification takes the same amount of time as with network layer rerouting, that is, a few milliseconds. All in all, the repair time is $T_{service} = T_{detect} + T_{notif}$ and is expected to be as fast as SONET [34].


next up previous contents
Next: Multicast fault recovery Up: Resilience and protection in Previous: Network layer protection   Contents
Yvan Pointurier 2002-08-11