The MPLS WG Archive

[Ajay Simha <asimha@cisco.com>]

On Sun, 2 Sep 2001, Hongwei wrote:

> Hi David,
>
> As much as I understand, MPLS technology is connection-oriented, and is
> below or at the same level of IP. Like all connection-oriented protocols,

I feel better using the term you used below and call it a "control driven"
process.  When we use LDP I'm not sure if we are "establishing a connection"
in the classic sense (of creating an end to end PVC).

-ajay

 > MPLS needs a set-up
procedure before traffic begins. PVC or SVC is
> established then. So called Control is the setup procedure(label requesting
> and label mapping in LDP and CR-LDP). It establishes the forwarding
> table,and it is before the traffic begins. So called Forwarding is in the
> traffic stage. The outgoing port is easily got from the forwarding table
> just because the setup procedure has established an explicit VC. So the
> traffic becomes easy. Label swapping is needed because the labels in the
> MPLS shim are locally significant, not globally(IP address is globally
> important). The locally important label can be dynamically allocated and
> released by the router. This adds flexibility. ATM also needs VCI/VPI
> swapping. If globally important identifiers such as (source IP, destination
> IP) pair are used as label of VC, multiple traffic between different
> processes in the source machine and different processes in the destination
> machine can't be distinguished.
>
> While IP has no setup procedure because it is a connectionless technology.
> So I just see its Forwarding function. At every hop, every IP packet has to
> find next hop's IP address and its corresponding MAC address by ARP.
>
> It is true that both IP and MPLS need consult the table, so MPLS forwarding
> is not necessarily faster than IP if packet's destination IP address can
> always match an entry in IP routing table. And now it is more apparent that
> MPLS is intended to combine the advantages of ATM and IP. MPLS can help to
> realize traffic engineering over IP networks, such as explicit routing can
> help allocate traffic evenly, and turn to alternative route quickly in case
> of network failure.
>
> I am just brave(rash?) enough to make above comments. I will appreciate any
> correction of my mistakes. THX in advance:)
>
> --Hongwei
>
>   -----Original Message-----
>   From: owner-mpls@UU.NET [mailto:owner-mpls@UU.NET]On Behalf Of David
> Escobar
>   Sent: 02 September 2001 18:32
>   To: mpls@UU.NET
>   Subject: Control and Forwarding functions
>
>
>   Hi:
>   It is said that MPLS makes a good separation between the Control and the
> Forwarding functions. It is also said that MPLS may use extensions of
> existing IP protocols to piggyback label distribution (MPLS-BGP,
> MPLS-RSVP-TUNNELS).
>   What is the meaning of good separation between the Control and the
> Forwarding functions? MPLS still uses the same Control protocols, just a
> little altered to provide label distribution by piggybacking and the
> Forwarding function still needs to make table look up to find the next hop.
> Even worse, it needs to make label swapping. Conventional IP also makes
> table look up but with the advantage of not requiring label swapping. Why
> can be inferred that conventional IP does not make a good separation of the
> Control and Forwarding function while MPLS does?
>

-- 
Ajay Simha
MPLS Deployment Engineer
IOS Technology Division
(919) 392-3141

"Study as if you were to live forever
 Live as if you were to die tomorrow"

 - Mahatma Gandhi