The MPLS WG Archive

[Seisho Yasukawa <yasukawa.seisho@lab.ntt.co.jp>]

Hello Vikram

>I have one more question regarding the draft. The following is from the
>draft.
>
>    Either the intermediate node or the leaf node may calculate the
>    multicast tree route in order to decentralize this calculation. So
>    all nodes belonging to the same multicast group need to have
>    information about the whole multicast tree topology in order to
>    prevent intersection of multicast tree.  Each node MUST NOT delete
>    the TERO subobjects that specify nodes on the explicitly routed path
>    in order to get the whole multicast tree topology information. Each
>    node MAY record the whole multicast tree topology from TERO. In this
>    way, all nodes can have the same information about the whole
>    multicast tree topology.
>
>The above states that all the nodes in the multicast tree have to have
>the full information about the multicast tree.
>I would like to know the technical reasons for this requirement.
>The obvious drawbacks with this requirement is the increase in the
>the message sizes with the increase in the number of nodes.

Technically, it is preferable to maintain the full information about the
multicast tree to avoid creating the multicast tree loop.
When considering leaf initiated Join process, the node receiving the
Join message become a leaf LSP setup initiator.
(We call this node as a Graft node)
Therefore the Graft node must be check whole multicast tree topology
when grafting leaf LSP to avoid the creation of multicast loop.

>In my opinion, if every node maintains the previous hop and the
>next hop(s) information, it should be enough, unless you see some
>specific technical reason for the above information. If it is
>only for the NMS show/information/display purpose, I feel it
>is better to do it at the source node itself and thus remove this
>requirement.

We also recognize the drawback of this architecture when considering
system implementation.
To maintain whole topology of multicast tree require huge memory
resource at each node.
This restricts network and system scalability.

Therefore, we want to modify this requirement mandatory to optional
in next version of our draft.

Thanks,

Seisho