The MPLS-OPS Archive

[Christopher Lewis <chrlewis@cisco.com>]

In Cisco routers, label allocation is downstream unsolicited, that is a LSR 
will define a label value for each known IP destination. In the 
connectionless IP world, this is most efficient.

In the case of an ATM LSR, labels are carried in the VPI/VCI field, which 
corresponds to a distinct ATM VC. VC space is limited in the connection 
oriented ATM world and consumes resources, so they are only allocated when 
they are needed (hence they are allocated on demand using DoD).

There are further considerations to ATM label allocation which makes the 
efficient use of VC space challenging.

As stated, each label corresponds to a VC, and we want to maximize the use 
of VCs. Consider an ingress LER that is an ATM switch. It has two IP 
streams coming in to it (on different interfaces) and both need to be sent 
to the one destination egress LER. If the ingress LER only has one label 
value for the destination egress LER, how would the egress LER know which 
of those cells to SAR together to form a packet? With just the VPI/VCI it 
does not have the information necessary for the extra step (SAR) that an 
ATM LER has to go through compared to an IP router that operates on packets 
in the first place. So, the label applied at the ingress LER needs to be 
assigned based on IP destination and incoming interface (incoming to the 
ingress LER in this case).

This requirement has the effect of increasing the number of labels an ATM 
LER needs to address the same number of destinations. To get around this 
problem, VC merge was created, which buffers cells together and only 
transmits them when a complete packets worth of cells has been received. 
This allows an ATM LER to use the same destination label value for ingress 
streams coming in on different interfaces that are destined for the same 
egress LER.

Chris

At 02:57 AM 10/25/2002, Gowda, Sidde wrote:
>Hi
>
>Label distribution can be done either in DOD or DU mode. It depends on the
>requirement. For eg. in case of L2 VPN (Martini) VCs are established using
>DU moode. DOD is not preferred in this case. Whenever VCID (L2 FEC) is
>configured, LSR sends a label mapping message to its peer. Since L2 FEC is
>not a part of routing information (neither destination address nor prefix)
>DOD is not useful. This is the one tipical example where Du is preferred.
>
>Siddu
>
>-----Original Message-----
>From: Prabhugouda Biradar [mailto:prabhu_b@huawei.com]
>Sent: Friday, October 25, 2002 12:44 PM
>To: mpls-ops@mplsrc.com
>Subject: [MPLS-OPS]: LDP DU Mode
>
>
>
>Dear All,
>
>Can some one tell me where and how DU mode of LDP is useful?
>I would like to know the actual deployment on the core.
>Are there any application of DU mode currently being deployed?
>Does any ISP is establishing LSP using DU mode of LDP?
>
>
>Most prominent applications of MPLS are VPN and TE.
>For TE you need to have CR-LSP (i.e. DOD). For VPN also
>Whenever you learn new PE router(by CLI or any other means)
>You can establish DOD LSP between the PE router.
>
>If only these are the two applications when exactly DU is configured???
>
>
>Thanks and Regards,
>Prabhu
>
>-------
>The MPLS-OPS Mailing List
>Subscribe/Unsubscribe:  http://www.mplsrc.com/mplsops.shtml
>Archive: http://www.mplsrc.com/mpls-ops_archive.shtml
>
>-------
>The MPLS-OPS Mailing List
>Subscribe/Unsubscribe:  http://www.mplsrc.com/mplsops.shtml
>Archive: http://www.mplsrc.com/mpls-ops_archive.shtml

-------
The MPLS-OPS Mailing List
Subscribe/Unsubscribe:  http://www.mplsrc.com/mplsops.shtml
Archive: http://www.mplsrc.com/mpls-ops_archive.shtml