Paths to MTU(topia?)
Hear me out
#### MTU propagation over EVPN Overlays
Use of ICMP multi-part extension to capture enough of the overlay-packet headers in PTBs so that tunnel interfaces can successfully translate underlay PTB into overlay PTBs. This is proposed in an IETF draft: [draft-saum-nvo3-mtu-propagation-over-evpn-overlays]
This seems _very_ promising to me
_Does_ put additional compute burden on TEPs, but seems relatively manageable
##### Purely Speculative Paths to MTU(topia?)
##### IP/ICMP PMTUD w/ inference?
* PMTUD works for _all_ IP traffic.
* Implemented at _IP/ICMP_ layer
* PLPMTUD _only_ requires participation by the endpoints
* Uses _inference_ of PMTU based on correlation of packet-length and _lack_ of acknowledgement
* Can we _infer_ PMTUD at the IP/ICMP layer?
* Yes, we can!
* We’ve all used the “ping –m do -s…” test to zero in on path MTU at _some_ point
* Let’s bake that behavior into host’s IP interface implementations?
##### Decouple Interface and Path MTU?
PMTUD (all varieties) are only designed to detect path MTU that is _lower_ than the local egress _interface_ MTU
But, as we know, _interface_ MTU is inherently untrustworthy.
Let’s “take the leash off” of path-MTU and allow it to probe up _past_ the local interface MTU value.
Would allow “senseless victim flows” to realize _actual_ path MTU, even though the origin network’s MTU had been “dumbed down” because of one weak-link somewhere upstream that only _some_ traffic was routed through.