The MPLS WG Archive

["Peter Ashwood-Smith" <petera@nortelnetworks.com>]

ashish wrote:
> 
> Hi Everybody,
> I have few queries regarding CR-LDP draft : draft-ietf-mpls-crldp-03.txt .
> 1. In traffic parameter TLV, there are 3 fields:
>   a. PBS : Peak Burst Size
>   b. CBS : Committed Burst Size
>   c. EBS : Excess Burst Size
>  Whats the relation between these .

   Bilel can answer these better than I but he's been off sick. 
I'll ask him to post something when he gets a chance.

> 2. In Resource Class TLV, its said there are 32 administrative groups or colors of Resources.
> Is there any document or draft explaining these Resource Classes.
> I have checked
> a)cr-ldp draft and
> b)RFC -2702 - requirements for Traffic Enggineering over MPLS.
> But there is not much detail regarding Resource Classes and how Resources are classified into 32 Resource Classes.
> Are these Resource Classes Implementation Dependent or there are some standards for these.

   The resource classes are intended to be a generic set of properties that can be compared against a
generic set of constraints. The idea is that you do not need to know (on a global basis) what these 
individual bits mean. Within an administrative domain, the network operator will decide what 
each bit means and assign them accordingly. 

   For example: 

       Let's say you want to assign bit #4 to mean "protected link". You would then go round
your network and when provisioning links which have some kind of physical protection, you would
set bit $4 in the Resource Class vector. This would then cause this bit to be propogated around.

       Now, when you want to establish a CR-LSP which only traverses "protected links" you simply
remove all links which do not have bit#4 set, in addition to those that do not have the required
bandwidth, and then run Dijkstra over the remaining topology.

       If one of your neighbouring systems uses a different bit to mean "protected" then either
you have to map one to the other, or, simply not honor his request.

    Aside:

        We did actually propose having common meanings for 1/2 the bit vector so that it would
be less likely that we would have boundary inconsistencies later but we were beaten into
submission ;)

    Cheers,

    Peter Ashwood-Smith