Re: Multicast in BGP/MPLS VPN with pseudo wires technology

[xuxiaohu_41208%HUAWEI at notesmail.huawei.com.cn]

Dear Rosen,

Thanks a lot for your reply. I fully agree with your viewpoints.

Just as you said, "when the packet is an MPLS packet, it may be carrying a
label which
identifies a particular tree,  rather  than just  identifying  a particular
<S,G>.  In  that case,  you would not  need to  know the input  interface
in
order to figure out the tree  on which the packet is traveling."

But how to describe a tree? I think the root and the instance are two
necessary components. The root of the tree is just the ingress PE, the
instance of the tree can be represented by MVPN instance, which is just the
meaning of the inner label describled in my draft. Of course, here only
refer to the scenario of one tree per MVPN. As to describing the MPLS label
as virtual interface, it's just for the purpose of understanding easily as
the ordinary process of PIM is familiar to all of us.

Regarding bidirectional issue, in the context of my draft, both the
multicast control message and multicast data traffic are travelled over
PWs. As you mentioned, bidirectional feature can be used for sending
control messages.

The URL for my draft is:
http://www.ietf.org/internet-drafts/draft-xu-l3vpn-2547bis-mcast-01.txt

Any comment is welcomed.


Sincerely

Xiaohu Xu






Eric Rosen <erosen at cisco.com>@ietf.org 于 2005-10-25 01:03:49

请答复 给 erosen at cisco.com

发件人:     l3vpn-bounces at ietf.org


收件人:     xuxiaohu_41208%HUAWEI at notesmail.huawei.com.cn
抄送:  cli at futurewei.com, l3vpn at ietf.org, sdawkins at futurewei.com

主题:  Re: Multicast in BGP/MPLS VPN with pseudo wires technology


> In the context of my draft, the pseudowires concept only means a
> bidirectional virtual circuit used as a MVRF related virtual interface.

I don't think it needs to be  bidirectional, unless we are also using it
for
sending control messages.

> RPF checking is a very important mechanism for multicast processing.

Sometimes in multicast  processing, when you receive a  data packet you
need
to figure out the tree on which the packet is traveling.  When the packet
is
an IP  packet, all  the information  you have is  the <S,  G> and  the
input
interface.  The  packet carries  no identifier  of the tree  on which  it
is
traveling, this  must be inferred from  the <S,G> and the  in put
interface.
However, when the packet is an MPLS packet, it may be carrying a label
which
identifies  a particular  tree, rather  than just  identifying  a
particular
<S,G>.  In  that case,  you would not  need to  know the input  interface
in
order to figure out the tree  on which the packet is traveling.  Therefore
I
am not really sure we will need this "virtual interface" concept.

I think we will need to pin  down the details of the control plane(s)
before
we can fully evaluate this issue.