Hi Robert,
On 19/07/2026 02:05, Robert Raszuk wrote:
Hi Peter,
I read your draft with interest. Have two questions:
#1
From the text it seems that classification on what you are offloading
is purely destination based. If so I find it way too coarse. How about
at least selecting src-dst based offloading ? If not going to all 5
tuples?
IGPs natively compute and install destination based prefixes in RIB. You
would need some PBR based involvement to make it more than that. Nothing
stops an implementation to do that, it's all local behavior on a router
doing the DCM offloading.
#2
When N nodes doing offloading onto the same upstream link Lx how do
you avoid in a fully distributed way a congestion on that link Lx.
Note that the Lx link could be Z nodes away.
For N nodes to do offloading it would require congestion on all of those
N nodes.
If that happens, and Lx utilization hits the high-utilization mark, it
will signal that in form of the high-utilization affinity and all N
nodes will stop adding more offloaded traffic over it, while keeping
what they have done already.
More details are in section 11.1, the second-to-last paragraph in
particular is of interest in this scenario.
thanks,
Peter
Thx,
Robert
_______________________________________________
Lsr mailing list -- [email protected]
To unsubscribe send an email to [email protected]
