next up previous contents
Next: Directions for future work Up: Conclusion Previous: Conclusion   Contents

Contributions

In Chapter 3, we presented a graph algorithm which computes a single backup path to protect a multicast routing tree. The backup path is computed after the multicast routing tree establishment and before a link failure occurs, making it suitable for pre-planned rerouting mechanisms. The aim of the algorithm is to minimize the number of members of the multicast group dropped from the communication when a single link fails. We showed how a backup path determined by the algorithm could be used to reroute traffic so that no node is dropped from the tree when a single link of the protected path fails. We determined the complexity of the algorithm in the average and the worst case. We also gave extensions to the algorithm to support dynamic multicast groups where nodes can join and leave after the establishment of the communication.

In Chapter 4, we presented MPLS multicast Fast Reroute, a pre-planned rerouting mechanism that can use the backup path computed in Chapter 3 to protect a multicast routing tree when a single link fails. MPLS multicast Fast Reroute uses a probing mechanism to detect link failures. The nodes that detect the failure then propagate link failure notifications over the multicast routing tree. Two routers, one at each end of the backup path, switch traffic over the backup path when they are notified of the link failure. When the link is repaired, the nodes that detect the failure also detect the link repair and propagate link recovery information on the multicast routing tree. The mechanisms used to detect the repair and notify the routers of the tree of the link repair are the same as those used when a link fails. The two routers which switched traffic over the backup path perform switchback by stopping forwarding traffic on the backup path. During switchover, nodes downstream of the failed link with regards to the center of the tree are temporarily disconnected from the multicast routing tree. During switchback, certain links of the tree see an increase of traffic and certain nodes of the tree receive duplicate packets.

In Chapter 5, we presented our implementation of multicast extensions to MPLS-Linux. MPLS-Linux is a unicast MPLS implementation that runs on Linux PCs. To our knowledge, our implementation is the first non-proprietary MPLS multicast implementation. We also presented the MulTreeLDP protocol which implements the MPLS multicast Fast Reroute mechanism described in Chapter 4. We evaluated the performance of multicast MPLS-Linux and MPLS multicast Fast Reroute. Overall, multicast MPLS-Linux is able to forward fast Ethernet traffic on PC routers with no dedicated hardware. MPLS multicast Fast Reroute can repair a multicast routing tree in a few tens of milliseconds which mostly correspond to the time to detect the failure. Because of the timer resolution limitation on the Linux operating system, it is not possible to build faster software link failure detector. When the failed link is repaired, MPLS multicast Fast Reroute can switch the traffic back on the original multicast routing tree in a few milliseconds.


next up previous contents
Next: Directions for future work Up: Conclusion Previous: Conclusion   Contents
Yvan Pointurier 2002-08-11