IETF90 Toronto, 22 July 2014   To-Do item for Eric Vyncke on draft-ietf-opsec-lla-only: Wes expressed concern that even though the document doesn’t read like a recommendations document, it might seem that it has WG support.  The IESG had discussed, and Joel Jaeggli said he needed the authors to add a disclaimer saying that this isn't recommended by IETF or something similar. ------------------------------------------------------------------ Topic: Implementing RPKI-based origin validation  two countries at a time: The Ecuadorian and Costa Rican case studies Speaker: Sofia Silva ------------------------------------------------------------------ Draft-fmejia-opsec-origin-a-country-00 (describes on possible deployment strategy for origin validation.) http://tools.ietf.org/agenda/90/slides/slides-90-opsec-0.pdf Slides and presentation present a good set of lessons learned and practices around rolling out origin validation. * Randy Bush: re: L2 IXP, refer to work done in NZ using a NetFlow-enabled switch driven by RPKI-aware route servers, where it prevented forwarding for invalid prefixes.  Previous work had suggested data plane didn’t trust control plane.  Do you have any data about dropping Invalids?  E.g., NOC ticket. Sofia: I don’t, but Fabián might. * Wes George: Hosted model within RIR has a limitation regarding downstream customers.  Maybe add something to the draft about whether downstream customers need a relationship with RIR for their delegated prefixes, in the hosted model.  Sofia: Dev team is working on the implementation.  Providers create ROAs for customers, because if they don’t have PI space, they can’t log into LACNIC. * Wes:  Does LACNIC have a published SLA for hosted system? Sofia: Not sure.  But it’s a production system, so they should. * Doug Montgomery:  Progress on number of valids and percentage of address space is great.  Do you offer any support on tracking down Invalids? Sofia: We’re doing something like that.  Alejandro Acosta is measuring Invalids, with an idea to contact those with invalid ROAs and ask them if they made a mistake or hijacked. * Randy Bush: Most common case seems to be issuing a ROA with max length of 16 but then announces 24s.  If they still announce the /16, then the /24s get dropped and they only lose their traffic engineering.  There used to be significant problems among some operators in RIPE, but we shamed them and they fixed it.  So, loss of TE costs money.   ------------------------------------------------------------------  Topic: draft-gont-opsec-ipv6-eh-filtering-00           Fernando Gont, 15 min Speaker: Requesting author did not appear ------------------------------------------------------------------ Discussion: * Eric Vyncke: Another work in progress by Campanikis(?)  Handling of hop-by-hop, document suggests drop, but I think you could also chose to ignore or rate limit. * Jen Linkova: I agree.  Draft says, “If you filter this, you’re going to break A, B, C; recommendation: filter.”  Don’t like making default policy Drop.  Let operator decide.  Make administrator explicitly decide. * Ole Troan: in 6man we have  a STD RFC7045 which this document overlaps. * Joel:  Yeah.  Let’s not contradict recent recommendations we made.  Our opinions can change, but not this fast. * Jen Linkova:  For Router Header, says drop RH0, but at least some other cases.  No mention of ongoing work.  For Experimental, suggests log-and-drop or ignore-and-forward; should be 4 permutation. * Mikhail Abramssson: Sometimes says “should not filter,” other times says “should pass.”  Is there a difference?  Is there a difference in behavior between core node and edge nodes? Maybe edge should have different policies. ------------------------------------------------------------------  draft-gont-opsec-ipv6-firewall-reqs-01           Fernando Gont, 15 Min draft-gont-opsec-ipv6-firewall-reqs ------------------------------------------------------------------ * Vyncke: (unable to hear due to cross-talk) * Joel:  It’s a -00, premature to discuss with nobody here to listen to comments.