Hi Huaimo,

Thank you for your discussion.  Unfortunately, I (and apparently others) are 
still not following you. We need you to be very specific and explicit because 
if your algorithm is accepted, we must write code that implements it. It will 
not help interoperability if we have to guess about what you meant.


> For example, for the failure of a link between node A and node B on the FT, 
> the backup path for the link computed is deterministic and the same even 
> though there are many links attached to every node. The start point and end 
> point of the backup path are deterministic, and are the node with smaller ID 
> and the node with larger ID respectively.


Ok, I’m already quite confused. Which sets of nodes are we considering when 
looking at the node IDs?  Do you mean:

a) Only nodes A and B.

Without loss of generality, let’s assume that node A has the smaller ID.  Are 
you suggesting that we compute the least hop count path from A to B?  What 
happens if there are links on that path that are already on the 
FT? If adjacency CD also straddles the partition, wouldn’t that be a more 
optimal solution?  An AB backup path might require several hops.

In many cases, we got to a partition because of a double failure.  Link EF also 
failed to create this situation. How do all nodes know to compute the AB path 
rather than the EF path? We cannot depend on time, since LSP propagation is not 
deterministic.  Some nodes may see the EF failure first, others may see the AB 
failure first.

b) The smallest ID in one partition (A) and the largest ID in the other 
partition (B).

This raises similar questions: A and B could be anywhere within their 
respective partitions.  How is the shortest hop count path between them optimal?

c) Something else entirely?


> The backup path is the path with minimum number of hops from the start point 
> to the end point, wherein whose links are not on the FT. When there  are 
> multiple paths with the same minimum number of hops, only one of them is 
> selected deterministically.


How do we select the path uniquely, and in a distributed fashion?  I can see 
many potential approaches, but I’m unclear on what your intent is.

Tony


_______________________________________________
Lsr mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/lsr

Reply via email to