Re: [mpls] MPLS-RT review of draft-bonica-mpls-self-ping

[Ronald Bonica <[e-mail]>]

Mach,

In a previous email, you said:

> 4. LSP Ping and BFD (especially S-BFD) are alternative tools for LSP readiness
> detection. If the scalability of control plane is the concern, there should be an
> analyses whether S-BFD cannot satisfy the requirement.

The following text addresses that issue.

                                         Ron


Text
====================================
Consider the following use case: 

- A network contains 100's of iLSRs
- Each iLSR is also an eLSR
- The network contains a full-mesh of RSVP-signaled LSPs (tens of thousands of LSPs)
- A network event occurs, causing half of those LSPs to re-signal
- RSVP churns, CPU's boil.....

Now, each iLSR needs to verify the readiness of its LSPs before attempting to forward traffic through them. The following are requirements of the mechanism used to verify readiness:

- it must return results quickly. (There is no time to set up session state or visit the control plane on the egress).
- it can consume only minimal control plane resources at the ingress
- it cannot consume any control plane resources whatsoever on the egress

The last requirement is very important. If the verification mechanism burns control plane resource on the egress, where CPUs are already saturated, network convergence time will suffer.

Only LSP Self ping fulfills the final requirement. Beyond that, LSP Self ping terminates as soon as it has verified LSP readiness.