[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: A philosophical question about FA and FA-LSP

hi - see in-line:

Kireeti Kompella wrote:

On Sat, 5 Mar 2005, dimitri papadimitriou wrote:


What's special?  Nothing really.  A "normal" TE link usually has a
(single) physical interface associated with it; an FA TE link is
usually multi-hop.  A "normal" TE link usually is accompanied by an
IGP adjacency (at least, in the packet domain); an FA TE link usually
doesn't.


but if a couple this with your answer "Correct -- LSPs created in IGP A
can be advertised as TE links in IGP B. This can be used (say) for
inter-area signaling." don't you have now an adjacency accompanying the
TE link advertisement in IGP B



Not necessarily.  Say you have:

       IGP B - X == IGP A == Y - IGP B

X and Y participate in both IGPs; X has an LSP to Y, and advertises it
into IGP B.  All's fine.

I'm guessing you are referring to the fact that IGP B appears
partitioned?  Well, to connect up the piece on the right with the
piece on the left, one can:
 a) create an adjacency over the FA LSP (only possible if LSP is PSC)

ok but is that not another adjacency accompanying the TE link advertisement for (or in) IGP B as stated in my previous reply ?

 b) have a virtual link (OSPF) from X to Y
 c) have some other "backdoor" connection between the two parts.


the same applying for the other cases

- therefore why is there a need to rename
the canonical inheritance i.e. why not just speak in that case about a
TE link or a "hierarchical TE link" as it was the case in one of the
earlier versions of the LSP-HIER document and keep the FA concept when
dealing with a unique instance ?


You can call it what you like :-)  "FA" in particular refers to the
idea that the "adjacency" is primarily for forwarding, and not
necessarily for the control plane.  However, the latter is not ruled
out (as in case (a) above.)

initially the term has been introduced to implicitly allow for the creation of a forwarding adjacency *without introducing a control plane adjacency* solving the facto scaling of the control plane both in terms of routing (creation of a data plane adjacency does not trigger creation of a routing adjacency therefore the number of routing adjacencies remains constant independently of the number of data plane adjacencies) and signaling (resulting from the LSP nesting operations), and the inherent problem associated to the so-called "unknown adjacency" problem that was also the achille heel of the IPoverATM approach where one had to create a full-mesh of VCs to determine the reachable end-points from the IP level

i do believe we are in agreement on this point - if not please let me know - now on the issue concerning not ruling out case a) as described here above how do you position this with the following paragraph in the LSP-Hierarchy document

"   It is expected that FAs will not be used for establishing ISIS/OSPF
   peering relation between the routers at the ends of the adjacency."

probably we do not have to preclude this but it would be advisable to consider a self-consistent terminology to describe the sub-cases we are dealing with

thanks,
- dimitri.

Kireeti.
-------

.