The MPLS WG Archive

[Alia Atlas <aatlas@avici.com>]

At 10:32 AM 11/20/2002 -0500, Thomas D. Nadeau wrote:
>At 10:07 AM 11/20/2002 -0500, Alia Atlas wrote:
>>At 09:58 AM 11/20/2002 -0500, George Swallow wrote:
>>> > The concept of avoiding the reserved labels in an MPLS hash as a 
>>> "best way"
>>> > makes some sense, though it would probably only be gradually available as
>>> > hardware changes.
>>>
>>>One should observe that the technique is not useful unless you are
>>>100% sure that very box in the path from source to destination is
>>>behaving this way.
>>
>>Why is it not incrementally useful?  Every LSR along the path that does 
>>this decreases the cases where an OAM packet may take a different route 
>>from the data packet.   It does require every LSR to implement to 
>>completely avoid OAM packets from taking different routes, but 
>>incrementally does simplify determining what happened to the OAM packets 
>>as compared to the data plane packets because there are fewer potential 
>>branch points.
>
>         There are cases where all LSRs need to understand
>replies via the control plane in LSP Ping.  This may not
>work unless everyone along the path understands what to
>do.

Tom,

Now I am confused.  We are discussing the load-balancing done internally in 
an LSR.  This is currently proprietary and different between different 
LSRs.  How does recommending not including the OAM labels suddenly break 
MPLS ping?  This has nothing to do with understanding the replies via the 
control plane in MPLS ping.

If it did, then we'd already have a problem with MPLS ping, because the 
load-balancing algorithms currently extant are different.

Alia