The MPLS WG Archive

[Curtis Villamizar <curtis@workhorse.fictitious.org>]

In message <6DEA508A9A0ED31192E80000F6CC176E2CA2DE@monk.datcon.co.uk>, Adrian F
arrel writes:
> Hi all,
> 
> Lou Berger suggested I poll the list to see what interest there is in being
> able to set up asymmetrical bi-directional LSPs without the need for an
> exterior signaling protocol.
> 
> In summary the initiator determines the need for a bi-directional LSP to the
> responder.  The TE component determines that (for various reasons - see
> below) the two directions of the LSP should follow different paths through
> the network.  This may involve different LSRs or simply different links
> between the LSRs.  How should the initiator set up the LSP?
> 
> Options seem to be
> 1. signal the forward path and use some external 
>    signaling method (such as SNMP) to request that
>    the responder signals the reverse path
> 2. over-load the MPLS signaling protocol to convey 
>    the request for the reverse path and its ERO to
>    the responder
> 3. enhance the signaling protocol to allow the 
>    two directions to be signaled from the initiator.
> 
> Obviously, 1 is simple (although perhaps a cop out - after all, we could use
> SNMP in place of RSVP-TE for uni-directional signaling!).  Option 2 has
> interworking implications.  Option 3 should only be considered if there is a
> real need.
> 
> A quick digression into motivation...
> 
> Call blocking can be a considerable problem when trying to set up
> bi-directional LSPs.  Simulations on various topologies of fully meshed
> optical switches show that call blocking is dramatically reduced when the
> selection of asymmetric bi-directional paths are allowed rather than only
> symmetric bi-directional paths.


Are you sure about this?  Is this the case when lights are out in one
direction but not the other in parts of the topology?


> Asymmetric optical paths can follow the same node hops in both directions
> (but using different links) or different node hops for each direction.
> 
> The general case of asymmetric bi-directional paths would have the initiator
> "set up" two separate LSPs and binding the two LSPs into one logical
> bi-directional path. The forward path would clearly be forward signaled
> (from the initiator). The reverse path would either be reverse signaled
> (from the responded) or forward signaled from the initiator, but in either
> case, the ERO would be calculated at the initiator.
> 
> A special case of asymmetric bi-directional path setup is when the forward
> and reverse node hops are identical, but the Links/Data Channels are
> different in each direction. It would obviously speed path establishment
> would be accomplished if both paths could be set up using one signaling
> exchange.
> 
> All response gladly received.
> 
> Regards,
> Adrian


A down side may be that asymmetric bi-directional paths will increase
the number of SRLGs that apply to a primary path and make it harder to
efficiently allocate backup paths.


Curtis