Re: [trill] Suggestion on RFC 6325 section 4.5.2, single parent ECMP link selection.

[Anoop Ghanwani <[e-mail]>]

On Wed, Aug 22, 2012 at 11:12 AM, Donald Eastlake <[e-mail]> wrote:
> Hi Ramkumar,
>
> On Wed, Aug 22, 2012 at 11:34 AM, Ramkumar Parameswaran (ramkumar)
> <[e-mail]> wrote:

>
>> When a node is connected to a single parent across several LAN links
>> which advertise a pseudonode (when pseudonode is not suppressed),
>> RFC 6325 section 4.5.1 already specifies a way to load balance
>> multi-destination traffic across these links by pulling a unique
>> link into each multi-destination tree.
>>
>> This is illustrated in the following example
>>
>>           |   Node B      | -> upstream, closer to tree root
>>           -----------------
>>              /  |   |  \
>>             |   |   |   |
>>             |01 02  |03 |04
>>             |   |   |   |
>>              \  |   |  /
>>          -----------------
>>          |   Node A      |
>>
>
>> Consider Node A and node B being connected to each other on 4
>> independent LAN links. Each of the four links above is a LAN link
>> with a DRB elected and pseudo-node advertised. The number to the
>> right of the link identifies the pseudo-node id operational on the
>> link. In this situation, per 4.5.1, since seven byte system id is
>> considered in load-balancing links across trees, other factors
>> holding, link 01 will be assigned to Tree 1, link 02 will be
>> assigned to tree 2 and so on.
>>
>> However, if the links were P2P links with 01, 02, 03, and 04 being
>> the extended circuit id of the node with higher system id (B, say),
>> then per 4.5.2, link 04 would be pulled into all multi-destination
>> trees, with no load-balancing on any of the other links.
>
> This provision in RFC 6325 didn't come out of nowhere. It was
> requested due, if I recall correctly, to limitation of some hardware
> in implementing the Reverse Path Forwarding Check for
> multi-destinaiton frames.

I do remember discussions around this issue, but can't remember
exactly what it was...I think it might have had to do with trying
to use these links for ECMP within the same tree.

Anoop