next up previous contents
Next: Hardware used for the Up: MPLS_trees Previous: The MulTreeLDP protocol   Contents


Experiments

In this chapter, we present the experiments that assess the performance of MPLS multicast Fast Reroute. All experiments are performed on Linux PCs equipped with Fast Ethernet network adapters. We successively evaluate the performance of all components of MPLS multicast Fast Reroute. These components are MPLS-Linux multicast, the link failure and recovery detection mechanisms, the link failure and recovery notification mechanisms, and the switchover and switchback mechanisms. In Section 6.1, we discuss the hardware used to conduct the experiments. In Section 6.2, we assess the performance of MPLS-Linux augmented with our multicast extensions by measuring the maximum throughput of data that can flow on a multicast LSP. We compare this maximum throughput with the throughput achieved with MPLS unicast and IP unicast. We show that MPLS-Linux with multicast support can fully exploit the capacity of Fast Ethernet links. In Section 6.3, we evaluate the link failure and recovery detection mechanisms. We simulate link failure and recovery by manually enabling up and disabling down an interface attached to a link of a mLSP. Disabling an interface is equivalent to physically cutting a link and enabling a disabled interface is equivalent to repairing a link. We measure the time required by the end nodes of the link to detect the simulated failures and repairs. We show that the time to detect a failure ($T_{fdetect}$) and the recovery ($T_{rdetect}$) of a link conforms to the distributions determined in Section 4.2, with an average value of a few tens of milliseconds. In Section 6.4, we measure the node notification delay $T_{nnotif}$ for notification messages. We show that the node notification delay is close to 1 millisecond and therefore the order of magnitude of the notification time $T_{notif}$ is a few milliseconds to tens of milliseconds depending on the multicast routing tree topology. In Section 6.5, we measure the total time to repair a multicast routing tree when a link fails. We send traffic over a mLSP and monitor the packet arrivals at the receivers. We measure the interruption time $T_{repair}$ seen by the receivers when we simulate a link failure. We compare $T_{repair}$ with the sum of $T_{fdetect}$ and $T_{notif}$ measured in the previous sections and show that the network is repaired in less than the SONET requirement of 50 ms.


Subsections
next up previous contents
Next: Hardware used for the Up: MPLS_trees Previous: The MulTreeLDP protocol   Contents
Yvan Pointurier 2002-08-11