The MPLS WG Archive

["David Allan" <dallan@nortelnetworks.com>]

Title: RE: I-D ACTION:draft-mo-mpls-protection-00.txt

Andrew wrote:

"If I read this correctly, there are only gains when the working and
protect paths incorporate a common node; although this could happen is
practice, it is not the preferred alignment."

Not entirely true, it is possible to conceive a link or node failure that affects a working path for one LSP and a protection path for another LSP. An argument could be made that an optimal and omnipotent CAC would go beyond simply adding working and protection reservations together and checking against the link capacity.

Whether this is a tractable compuation is another question entirely and I would suggest it is not: Such an algorithm would have to consider:

- individual LSP failure
- link failure
- node failure
- SRLG failure
for every resource in the network that had any association with LSPs that transited the local node, and have the CAC algorithm consider the worst case.

So one complication is that for every LSP (working or protection) that transits the local LSR, the computation needs to know all details about the protection counterpart (protection or working respectively). How can it know this? Is there not a massive synchronization problem if more than one protected LSP is being simultaneously routed across the network. (can't compute what has not completed routing yet).

I would have to be convinced that "a whole lot of bandwidth" miraculously appeared as a result of this optimization before I considered mechanisms to ensure I could reasonably and usefully compute this optimization. My suspicion is that unless I am doing 1:1 global repair on "everything" (which is not somthing I would do), the bandwidth "just isn't there" and may not be there anyway.

later
Dave