I synced up with PK in-person this morning and clarified his questions. I'm
reproducing my response here on the list to keep record.

The weights are specified per sub-candidate path.

If the sub-candidate path is dynamic, the specified weight is distributed
across the set of segment-lists as dynamically computed by the
compute-engine. So for example, if the specified weight for the dynamic
sub-candidate path is 3 and the computation result yielded 3 equal cost
segment-lists, then the weight associated with each segment-list would be
1.

If the sub-candidate path is explicit, the specified weight translates to
the weight of the corresponding segment-list.

Regards,
-Pavan

On Wed, Jul 24, 2019 at 1:23 PM Przemyslaw Krol <[email protected]> wrote:

> Hi Vishnu,
>
> Clarification question, if I may.
> Are you envisioning weights to be normalized for all sub-candidate-paths
> (regardless of the candidate-path they belong to), or rather propose to
> have a 2-level split:
> - weight for candidate-path grouping specific sub-candidate-paths defining
> "global" share of traffic, in comparison with other candidate-paths
> - weight for specific sub-candidate-path defining "local" share of the
> traffic within specific candidate-path, in comparison with other
> sub-candidate-paths in that particular candidate-path
>
> thanks,
> pk
>
> On Wed, Jul 24, 2019 at 6:42 AM Vishnu Pavan Beeram <
> [email protected]> wrote:
>
>> Authors, Hi!
>>
>> There are some use-cases where the candidate-path (multipath) needs to be
>> constructed in such a way that a part of the multipath (a set of
>> segment-lists) uses one set of constraints, while the other part (another
>> set of segment-lists) uses another set of constraints. Consider the
>> scenario where the traffic needs to be steered onto a policy in such a way
>> that a specified portion of it uses paths that traverse only blue links
>> while the rest uses paths that traverse only red links.
>>
>> The current semantics of a candidate-path as defined in Section 2.2 of
>> version 3 preclude such use-cases. It should be possible to let the
>> semantics be a tad more flexible than they currently are and cater to the
>> above scenario (and the likes of it).
>>
>> There are a few different ways of addressing this, but I would like to
>> propose one that seems least disruptive.
>>
>> In addition to the currently specified semantics:
>> -- A candidate path may comprise of a set of sub-candidate paths where
>> each each sub-candidate path is either dynamic or explicit. The
>> sub-candidate path is the unit of signaling of an SR policy between a
>> headend and a controller..
>>
>> Regards,
>> -Pavan
>> _______________________________________________
>> spring mailing list
>> [email protected]
>> https://www.ietf.org/mailman/listinfo/spring
>>
>
>
> --
> Przemyslaw "PK" Krol |  Strategic Network Engineer ing | [email protected]
>
_______________________________________________
spring mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/spring

Reply via email to