The MPLS WG Archive

["Sukanta Ganguly" <sganguly@opulentsystems.com>]

Eric,
  I do agree that it does not completely fit into L2 but even then it does perform the act of maintaining the connection orientedness. The MAC-to-MAC analogy can be applied to the MPLS realm too. If there was no MAC then between routers there would be the MPLS labels which would uniquely identify the ports/interface onto which the packet would have to go. Since we still depend on the MAC-to-MAC at the Ethernet layer, we do overlook this aspect. Thinking more into it, if we slice of the MAC layer underneath and rely on the MPLS labels to be a representation of the ports then do we think it will work? I am sure I can explain its working. 

SG

*********** REPLY SEPARATOR  ***********

On 9/4/2001 at 6:07 PM Eric Gray wrote:

>Brijesh,
>
>    A connection oriented service is one that maintains a persistent
>connection between two stations. Connection oriented services
>require connected stations to be constantly talking to each other,
>regardless of whether or not they have anything worth saying.
>
>    Not all MPLS LSPs are between two stations.  Topology based LSPs
>(e.g. - LDP or BGP established) form subsets of the paths from multiple
>sources to some definition of a destination.  In this case, LSPs may
>capriciously start and stop at any point along such paths and may even
>skip around, taking a different path, based on a short term change in
>routing.  Also, in these cases, talking between stations required to
>maintain LSPs is peer to peer, rather than LSP end to end.
>
>    Also, while MPLS labels may be carried as part of a L2 header,
>MPLS is not an L2 mechanism in the sense of the OSI model.  It
>exists between the Data Link Layer and the Network Layer.  One
>can argue that it 'fits' this model between layers or one can argue
>that it doesn't fit - but either position is not very important.  After
>all, a model is only a model.  It is - at best - a subset of reality.
>Often a model's most useful feature is that it allows us to discuss
>some complex notions (e.g. - a network) in a way that is not hopelessly
>confusing.
>
>--
>Eric Gray
>
>You wrote:
>
>> I don't know why some people will not accept that MPLS
>> is mainly a connection oriented technology. The main
>> characteristic of a Connection Oriented service is the
>> long term association required between two end points.
>> The process of how such association is established is
>> not important. It is also not important if the
>> connection doesn't begin or end at some hosts (why we
>> need to use ISDN or X.25 definitions of connections?).
>>
>> It is quite obvious that there is very clearly
>> established permanent association between two end points
>> of LSPs. All LSRs also maintain the necessary states
>> about particular paths (similar to circuit switched
>> network).
>>
>> From all this, MPLS clearly is the connection oriented
>> layer at layer 2. Of course, I have also seen comments
>> that MPLS doesn't fit in ISO model. Oh Really, !.
>>
>>
>> Cheers,
>>
>> --brijesh
>> Corona Networks
>> > OK, here we go again....
>> >
>> > To its nature MPLS is not "connection oriented"! I also depends quite a
>> > bit on the definition of "connection oriented", but if we e.g. let an
>> > ISDN
>> > phone call be the model of a connection oriented technology, then you
>> > immediately see that MPLS misses the most important part of what a
>> > connection oriented technology is all about.
>> > ISDN is separated into two parts "call control" and "connection
>> > handling",
>> > together they form the connection orientation.
>> >
>> > So the "call control" handle e-2-2 compatibility and ability to take
>the
>> > call. This does not exist in MPLS. The call control also OK's the set
>> > up of the call and is once it is established dependent of the e-2-e
>> > connectivity, if the connectivity is broken the call is released.
>> > Nothing
>> > of this is present in MPLS.
>> >
>> > Sometimes one hear that MPLS is connection oriented because there is
>> > a mapping from node to node over the network, strictly this is not
>true,
>> > the only thing that is mapped incoming label to outgoing label and
>> > interface.
>> > The rest is routing in the standard IP connections way.
>> >
>> > As for the arguments that "most of the" LSP setup is connection
>> > oriented,
>> > usually meaning end-to-end (which really is edge-to-edge, there is not
>> > much idea to establish a connection edge-to-edge. But even so
>> >
>> > LDP DU independent liberal is not connection oriented
>> > LDP DU independent conservative is not connection oriented
>> > LDP DU ordered liberal is not connection oriented
>> > LDP DU ordered conservative is not connection oriented
>> >
>> > LDP DoD independent liberal is not connection oriented
>> > LDP DoD independent conservative is not connection oriented
>> > LDP DoD ordered liberal is not connection oriented
>> > LDP DoD ordered conservative could at best emulate part of a connection
>> >         oriented technology (signaling)
>> >
>> > RSV-TE mostly strictly follows IP routing and is not connection
>oriented
>> >        and if there are explicit routes it falls into the same category
>> >        as LDP DoD ordered conservative
>> >
>> > We are left with a technology where it is required to have an
>> > MPLS-external
>> > application that do the call control part of connection orientation.
>Why
>> > we would like to do that is beyond me, as we loses all the benefits of
>> > the
>> > IP and MPLS combination.
>> >
>> > /Loa
>> >
>> > --
>> > Loa Andersson
>> > Chief Architect,
>> > Utfors Research, Architecture and Future Lab (URAX)
>> > Utfors AB
>> > Råsundavägen 12
>> > Box 525, 169 29 Solna
>> > Office          +46 8 5270 2000
>> > Office direct   +46 8 5270 5038
>> > Mobile          +46 70 848 5038
>> > Email           loa.andersson@utfors.se
>> > WWW             www.utfors.se