Re: [mpls] entropy label indicator label in draft-kompella-mpls-entropy-label

[Curtis Villamizar <[e-mail]>]

In message <02ac01cb2d87$a1c83ac0$[e-mail]>
Yong Lucy writes:
>  
> Hi Kireeti and Shane,
>  
> Could we consider use one of reserved label for ELI purpose? This will make
> implementation much easy, i.e. not need to keep the state for each ELI
> label. The approach can be used in general for RSVP-TE and BGP as well.
>  
> Regards,
>  
> Lucy


Kireeti, Shane, Lucy,

I agree that a reserved label would be useful.

You should also remove the restriction that the entropy label is on
the bottom of stack.

With three more bits of information in the TE-LSDB an LSR can identify
a path over which it can carry a very large aggregate MPLS LSP
containing smaller MPLS-TP LSP.

  1.  Each LSR must identify itelf as "entropy label capable" (if it
      is capable, otherwise it remains silent on this).

  2.  Each LSR must identify itself as capable of limiting the stack
      depth over which it will hash on a per LSP basis if it is
      capable, otherwise it remains silent on this).

  3.  Each LAG or bundle must identify itself as multipath capable
      (implied in LAG, currently uncertain for bundle) and indicate
      the largest microflow it can handle (this would be the largest
      component link minus some epsilon if it was capable of moving
      traffic to accommodate a large microflow).  This is similar to
      (but not identical to) Maximum-LSP-Bandwidth in bundles.
      Perhaps this could be called Maximum-Microflow.

If this approach is used, the CSPF for an MPLS LSP containing MPLS-TP
requires pruning all LSR that do not have the first two capabilities
and pruning all links that do not state a Maximum-Microflow or that
state a Maximum-Microflow that is less than the largest MPLS-TP LSP
being carried.

Signaling must indicate the depth to hash to.  The ingress of the
aggregated LSP would put the entropy label for all sub-LSP that can be
subdivided into flows (generally most or all non-MPLS-TP LSP).

At each forwarding hop the the hash depth limit must be respected but
with the entropy label above that depth the balance could be
maintained without impacting MPLS-TP.

Note: There is no requirement to identify the large flows through an
explicity marking at ingress.  This comment is not relevant to this
draft, but I thought I'd point this out before the thread goes off in
that direction.  Please keep that discussion on a separate thread.

Curtis