The MPLS WG Archive

Cell Relay Retreat>MPLS WG Archive>month:2000-Oct> msg00202



[Date Prev][Date Next][Thread Prev][Thread Next]  
  [Date Index][Thread Index][Author Index][Subject Index]

Draft Minutes From Pittsburgh

  • From: Kireeti Kompella <kireeti@juniper.net>
  • Date: Tue, 17 Oct 2000 21:58:18 -0700 (PDT)
  • Cc: fhujber@hotmail.com, mpls@UU.NET

Hi Yangguang,

> One implication Frank hinted is that:
> 
> For A----B, even there could be many bundles between A and B, there needs only
> one control channel (no need for physical channel, a TCP session is enough)
> between them. Bundle in circuit network is different from "interface" in IP
> network in the sense that a bundle can be included in one LSA, but only send to
> neighbor once.

You have a good point.  In section 3.2 of the bundling document, it
says that a bundle must have a control channel, but is not very
clear that multiple bundles can share a control channel.  We will
make this clearer in the next revision.

Meanwhile, let me say this explicitly: between a pair of GLSRs (i.e.,
LSRs, OXCs, SXCs, ...), the requirement is that there must be at
least one control channel; the requirement is *not* one per bundle.
Corresponding to this control channel, one routing adjacency is
needed.

The routing module can generate LSAs for all the bundle(s) assigned
to a control channel when both the control channel adjacency is up
and the bundles are up (LMP can be used to say when the bundles are
up).  If a bundle changes, the LSA corresponding to that bundle alone
is re-flooded.

This addresses the issue of flood optimization: having multiple
bundles does *not* mean having multiple adjacencies, so flooding
optimization is not a requirement.

Kireeti.