Linda,
On 11/11/2021 15:33, Linda Dunbar wrote:
Since we didn’t have enough time at today’s LSR session, I want to
continue the discussion of draft-dunbar-lsr-5g-edge-compute on the
mailing list:
* Shraddha pointed out that advertising the “site aggregated cost” of
a Prefix (ANYCAST) might cause oscillation of traffic flow as other
egress routers with the same prefix attached can advertise different
metric to cause the change of the path computation.
OSPF-TE has the same problem, does it? Link Bandwidth or condition
changes advertisement can cause IGP path change?
IGPs do not use dynamic bandwidth as part of its calculation. TE is a
different beast, it has a way to pin the traffic to a certain path.
Shraddha has a valid point - we need to be careful with any metric that
changes often and can shift traffic suddenly. But if the metric is not
too dynamic maybe we can make it work.
What I'm not sure about is how would you prevent all existing sessions
to be shifted between the anycast locations if the new metric changes. I
would think you would want to keep the existing sessions where they are
and only use the alternate anycast location for the new sessions. If
that is the requirement, I'm not sure how we can achieve that purely at
the routing level as that would shift all the traffic.
From each node’s perspective, there has be a timer to adjust the IGP
path change with newly received “site-cost” metrics.
In addition, “Site-Cost” associated with the prefix is only considered
in computing the IGP path when the Flag in the FlexAlgo TLV is set to
indicate the need to include the “site-cost” metrics.
* draft-dunbar-lsr-5g-edge-compute is using
E-intra-Area-Prefix-LSA[RFC7684] and Prefix Reachability TLV to
advertise the “Site-Cost” associated with the prefix. At today’s
session, I heard that the “Stub Link” can be used as well. Want to
hear WG opinion on which one is more suitable?
stub-link in OSPF Router LSA is not extendible as it it not TLV based.
Any new data that you want to add to the OSPF link/prefix needs be
advertised as defined in RFC7684.
If you are referring to draft-wang-lsr-passive-interface-attribute-08,
authors have defined the new TLV using RFC7684 (not that I agree with
what it is meant to be used for).
One more point, the “Site-Cost” associated with a prefix is not only
intended as Tiebreaker. When the Flag in the FlexAlgo TLV is set to
indicate the need to include the “site-cost” metrics, the path
computation will include the value in the SPF computation.
It's up to us to define the best behavior to address the valid use case
we have.
thanks,
Peter
Thank you very much.
Linda
_______________________________________________
Lsr mailing list
Lsr@ietf.org
https://www.ietf.org/mailman/listinfo/lsr
