Hi Huaimo:

 

Replies are in line….prefaced with DA>

 

<snipped> 

1.      The alternate backup path would appear to also require the criteria of 
being link diverse with the FT if the goal is to protect against multiple 
failures.

 

[HC]: Can you give some more details about this?

[DA] There is a bit of a chain of logic I did not well elucidate…

If we have an FT sufficient to be complete in the presence of any single 
failure, AND if we have a multiple failures situation such that the FT has been 
partitioned, and the information at any node is incomplete, then IMO the 
heuristic to attempt a blind repair with the highest probability of success is 
to 

a.      Assume any observed failure is the worst possible class of failure 
(e.g. node, as if the FT is severed the surrounding nodes will only see one or 
some of the LSAs associated with the node failure).
b.      Attempt to restore using links that are not part of the FT as if I 
assume the probability of multiple failures decreases exponentially in 
proportion to the number of simultaneous failures, it has a higher probability 
of success….

On reflection ‘b’ seems too simplistic, and does not reflect that some 
knowledge of what parts of the FT survive in the partition the node 
contemplating restoration is in, would be available for decision making.  And 
the fact that the concept of path as a response to the failure scenario being 
discussed IMO is not realistic (I elaborate a bit below).

 

2.      If node failures are considered, I’m not sure what criteria is used to 
deem a backup path as useful…..

 

[HC]: Regarding to the failure of a node X on the FT, suppose that there are 
multiple (i.e., two or more) nodes that were connected to the failed node X 
through the links on the FT. For each pair of these multiple nodes, a backup 
path between this pair is computed and enabled for temporary flooding. Thus the 
backup paths will connect these multiple nodes on the FT, and the FT partition 
caused by multiple failures including the failure of node X is fixed through 
the backup paths for the failed node X and the backup paths for the other 
failures. 

 

For example, if the failed node X was connected to two nodes Ri and Rj (assume 
that Ri’s ID < Rj’s ID) by the links on the FT before node X fails, there is 
only one pair of nodes: (Ri, Rj). A unique backup path from Ri to Rj is 
computed and enabled for temporary flooding. This backup path will connect Ri 
and Rj on the FT and fix the FT partition caused by multiple failures with the 
backup paths for the other failures.

 

In another example, if the failed node X was connected to three nodes Ri, Rj 
and Rk (assume that Ri’s ID < Rj’s ID < Rk’s ID) by the links on the FT before 
node X fails, there are three pairs of nodes: (Ri, Rj), (Ri, Rk) and (Rj, Rk). 
A unique backup path from Ri to Rj, a unique backup path from Ri to Rk, and a 
unique backup path from Rj to Rk are computed and enabled for temporary 
flooding. These three backup paths will connect three nodes Ri, Rj and Rk on 
the FT, and fix the FT partition caused by multiple failures with the backup 
paths for the other failures.

 

DA> Again I need to back this up a bit and incorporate a bit more subsequent 
reflection in my response.  What I was referring to blurred two discussions, 
adding links in response to severing and your post where path establishment 
seemed to be based on a previously known network state.

 

As observed above, I do not think a restoration strategy focused on a repair 
path that assumes a link failure will do anything useful for the partitioning 
scenario under consideration.  I also do not see a simple heuristic for a 
collection of nodes that are blind to the overall state of the FT to create a 
new path in the FT as a distributed response and where no signaling is 
involved.  I’d assume that is why what is being discussed is to add links 
temporarily as that is about the only strategy that can work with unilateral 
decisions by single nodes not acting in a coordinated fashion….If a path is 
required, a node trying to instantiate a portion of the path cannot depend on 
its neighbor to independently come to the same conclusion, in fact for an 
actual repair the opposite is just about guaranteed.  

 

I hope that is clearer

Dave

Attachment: smime.p7s
Description: S/MIME cryptographic signature

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

Reply via email to