 |
 |
 |
 |
|
 |
|
 |
|
 |
|
 |
|
|
|
|
|
|
|
|
|
|
|
|
|
Dave, Thank you very much for the reply. What do you think of my suggestion? o
Is it possible to create a special BFD
session that all intermediate nodes will forward the BFD to all the paths in the
ECMP group? Whenever it happens, the node will mark a bit to the Discriminator
field. By receiving all the BFD frames with different Discriminator value, the
egress node can figure out how many alternative paths are between them. Linda Dunbar From: Dave Katz
[mailto:dkatz at juniper.net] On Nov 8, 2009, at 6:10 AM, Linda Dunbar wrote:
Dave
and David, Hope
you can help me to figure out the following issues with regard to
“draft-ietf-bfd-multihop-8”: ·
Multiple BFD sessions between one pair of
system was mentioned several times in the document. Is the purpose of having
multiple BFD sessions between the same pair for traversing different
alternative routes between the same pair? If yes, the document didn’t say how
to enforce it. Even the different Destination/Source addresses being used can’t
guarantee that those BFD sessions will traverse different paths. Is it out of
the scope of this document? The mention of multiple BFD sessions is primarily for
discussion about how to demultiplex such sessions, rather than to attempt to
deal with the liveness of alternate paths. As you point out, there is no
mechanism for binding a BFD session to a path, only to its endpoints, and even
that is somewhat fraught with subtlety (which is why the multipoint spec spends
most of its time talking about how to demux BFD packets.) As the whole point of BFD is to follow the same paths
that data is following, sharing fate as much as possible, trying to run BFD
over inactive alternate paths would be extremely difficult. Multiple *active* paths may exist (such as an LSP and
a routed path) and most of the verbiage is to provide a palette of mechanisms
that can be used to sort the BFD sessions out.
·
·
What are the benefits of enforcing BFD
sessions between a pair of nodes to traverse alternative paths between them? I
guess by doing so, the two nodes get to know how many different paths are
between them. Who will use this information? See above.
·
·
Your BFD base draft
(draft-ietf-bfd-base-09) does state that one of the goals of BFD is to traverse
all possible paths between two nodes. Does it include the alternative paths
between specific Ingress and Egress? I don't believe that it ever says such a thing, and it
isn't possible in any case. Can you point at the text that gave you this
impression?
·
·
Suggestion if you do think it is
necessary to traverse all the alternative paths between two nodes as indicated
in the BFD base document: o
Is it possible to create a special BFD
session that all intermediate nodes will forward the BFD to all the paths in the
ECMP group? Whenever it happens, the node will mark a bit to the Discriminator
field. By receiving all the BFD frames with different Discriminator value, the
egress node can figure out how many alternative paths are between them. See above.
o
Hope to
have more discussion in Linda
Dunbar Advanced
Technology Dept, Wireline Networks, Huawei
Technologies, Inc. |