The setup of the experiment is illustrated in Figure 6.8. We set up a multicast LSP of six nodes. PC1, PC5 and PC6 are the LERs of the tree. PC2, PC3 and PC4 are LSRs. The link between PC3 and PC4 is the unique link of the backup path. The PSLs are PC3 and PC4. The links between PC3 and PC2, and PC2 and PC4 are the links of the protected path. PC1 is a source and sends UDP packets of 8192 bytes at 40 Mbits/s on the tree. The receivers are PC5 and PC6. Monitor captures traffic on the link between PC2 and PC4 and on the link between PC4 and PC6 using tcpdump.
To measure the node notification delay , we simulate the failure and recovery of the link between PC3 and PC2. When we bring down interface eth3 of PC2, PC2 and PC3 detect a link failure. PC2 sends a link failure message to PC4. Upon reception of this link failure message, PC4 sends a link failure message to PC6. We use Monitor to measure . Since Monitor captures traffic on the link between PC2 and PC4 and on the link between PC4 and PC6, Monitor receives both notification messages and is able to compute the time difference between the instants at which it receives each message. When we bring up interface eth3 of PC2, PC2 and PC3 detect the link recovery. PC2 sends a links recovery notification message to PC4 and then PC4 sends a link recovery message to PC6. Monitor records the time at which it captures the notification message on the link between PC2 and PC4, the time at which it captures the notification message on the link between PC4 and PC6, and determines for the link recovery by computing the difference between the two recorded times. To bring up and bring down interface eth3 of PC2, we use the additional thread in MulTreeLDP we introduced in Section 6.3. This thread brings down and brings up eth3 at instants randomly chosen by the random number generator of the machine. After the interface is brought down and brought up 100 times, we stop and restart MulTreeLDP manually on all six machines. We repeat the experiment 25 times. Therefore we collect 25 series of 200 node notification delays.
Figure 6.9 shows how the 5000 samples are distributed. The average for is
=1.18 ms with a minimum of 0.127 ms, a maximum of 4.172 ms and a standard deviation of 0.37 ms. The notification time is a multiple of . In our experiment, there is at most one hop between the routers which detects the link failure and the PSLs thus and
. Therefore the average time to notify PSL PC4 of the link failure is 1.2 ms. With larger networks and larger trees, the value of is larger and the notification time can reach a few tens of milliseconds. It is possible to decrease by opening a TCP connection between every two neighbors in a multicast LSP at circuit establishment and closing the connection when the LSP is torn down instead of opening a new TCP connection each time a MulTreeLDP sends a message. This optimization is left for future work.