Today’s routers generally make themselves more- or less-attractive to
transit traffic through operator’s manipulation of their interfaces IGP
metrics or overload status. This all-or-nothing method lacks granularity
and does not take advantage of the wealth of connectivity and
health-check information readily available at the router.
We propose the notion of “connectedness value”, and describe an approach
that allows a router to take pre-configured actions depending on its
current level of connectedness to the rest of the network. The
“connectedness value” is derived from not only reachability information,
but also from OAM data and other policy-based criteria. For each
router, an operator can now define what “well-connected” means, and what
should happen should the router becomes less than “well connected".
In this presentation we will discuss the concepts around defining
and acting on connectedness value, we well as the prototyping we have
done for several use-cases relevant to a content-provider network with
lots of remote POP/cache facilities. We will seek feedback from the
community on the usability of the proposed approach, as well as ideas
for enhancements.
https://www.nanog.org/sites/default/files/Hoffman_Suffering_Withdrawal.pdf
No comments:
Post a Comment