The MPLS WG Archive

[neil.2.harrison@bt.com]

Kireeti...what a lovely demonstration of the meaningless L1, L2, L3
classification.  The only classification that makes any sense is the
modal one, ie co-cs, co-ps and cl-ps.....and then an understanding that
these modes are not the same, so any '1 size fits all' treatment across
all modes makes no sense.  Indeed, it is the differences that are the
important bits......not sure many folks get this however.  All these
modes are 'peer L3 networks' in the unreal and restricted OSI 'L3
network layer' view of the networking world.

BTW - There are no QoS classes in the co-cs mode.....this is one of its
key defining differences.  Some others are:
-	control/management must be OOB....this is not an option in this
mode
-	can't violate the connectivity constraints of the co mode, ie
merging is simply not possible here
-	one always has a fixed/known hierarchy in the data-plane...so
one can force functional decoupling between layer networks.
All of these are very important properties of the co-cs mode......you
don't find them *forced* in the cl-ps mode, and because of this one can
sure screw them up in the co-ps mode.  If anyone is wondering 'why' this
is, it is down to the nature of how the link-connections are
defined/manifested in the co-cs mode, ie here they must take on a
real/physical property of either time, space or frequency and the
bit-rate must be fixed at a known value (cf the co-ps mode where the
link-connection IDs are simply abstract 'numbers').

Further:
-	given a network layer N inherits the performance of the layer
N-1 network (ie link-connections in layer N are provided by trails in
Layer N-1....theis is the fundamental property (in G.805) that defines
layer N topology creation);
-	this is a recursive behaviour to the duct;
-	there are (currently at least) no rules for governing what
client/server relationship can exist between pairs of modes (there are 9
permutations);
-	there are no rules govening *how many* client/server instances
can exist in an instantiation of a network arch that delivers some
end-system application (ie right at the top of the stack);
.......then its clear that trying to define any end-end network
performance objectives and then trying to apportion these to partitions
of this (eg operator domains) in any meaningful/fair manner seems an
impossible task at the current time.  Thus, any talk of 'QoS' is also a
rather meaningless exercise.

BTW - If folks were more familar with functional architecture (ie
G.805/809 stuff) all the above would be obvious....I saw some of the
later posts and its clear to me folks are very confused about layer
networks.  Very seriously....please have a look at chapter 3 of
Reid/Sexton ('BB networking') as this will help.

regards, Neil

> -----Original Message-----
> From: mpls-bounces@lists.ietf.org 
> [mailto:mpls-bounces@lists.ietf.org] On Behalf Of Kireeti Kompella
> Sent: 12 November 2004 21:00
> To: SP
> Cc: mpls@ietf.org
> Subject: Re: [mpls] what routing protocols have what to do with QoS?
> 
> 
> On Thu, 11 Nov 2004, SP wrote:
> 
> > since MPLS over ip over ethernet...
> > and since ip over ethernet (layer 3 over layer 2)...
> 
> Let MPLS be Layer N.
> 
> By the above, N > 3.
> 
> You can also carry Ethernet over MPLS,
> 
> Hence, 2 > N.
> 
> Therefore, 2 > 3.  QED.
> 
> Hope that helps :-)
> 
> Kireeti.
> -------
> 
> _______________________________________________
> mpls mailing list
> mpls@lists.ietf.org
> https://www1.ietf.org/mailman/listinfo/mpls
> 

_______________________________________________
mpls mailing list
mpls@lists.ietf.org
https://www1.ietf.org/mailman/listinfo/mpls