|
Hi Yangguang,
The protection path is set up (e.g., path computation and label mapping) together with the working path when the connection request arrives. However, protection paths are activated and put into service after the network fault, assuming shared protection. With respect to the setting up of the protection path, we do not change the end-to-end signaling using RSVP-TE or CR-LDP. Only the activation process is done through flooding. The rationale behind using end-to-end signaling for the setting up of the protection path is that this is a process that is not bound by a certain time constraint.
At activation time, assuming a WDM network, ingress-initiated end-to-end signaling does not scale well and cannot guarantee time bounds.
The scalability issue comes into the picture because for every wavelength in a fiber, a notification message would be needed using end-to-end signaling.
The issue with time bounds is the amount of messages that are sent out. Let us assume a fiber cut disrupted 128 wavelengths. Then, the 128th notification message from the detecting NE will have to wait in the outgoing queue of that NE until all other 127 messages have been sent to their respective ingress LSRs. At every NE in the network, again, in a WDM network, each NE may receive again from 0 up to 128 notification signals to activate the protection path (including reconfiguration of the switching core). Flooding requires less messages in the network and ensures that only one message has to be processed at any NE, solving the issue of buffering at every NE.
In addition, the ability to do the reconfiguration in parallel (if supported by the hardware) is allowed through flooding, but not allowed by end-to-end signaling, since the signal may only tell of one LSP activation.
You are correct in your assumption that each NE has the knowledge of which path it is protecting. This is done during the path setup phase.
-----Original Message-----
Richard,
I was reading your "Fault Notification and Service Recovery Protocol" draft. It's a very nice one.
Several questions for me to fully understand your reasoning for flooding the fault notifications. My understanding of your proposal is that it doesn't need an end-to-end signaling to set up the protecting path. Rather, the protecting path is set up segment by segment triggered by the flooded notification messages. Am I right? Also are you assume that each NE has the knowledge of which path it is protecting?
Yangguang |