Minutes IETF104: pim
|Meeting Minutes||Protocols for IP Multicast (pim) WG Snapshot|
|Title||Minutes IETF104: pim|
|Other versions||plain text|
IETF 104 pim pim-dr-improvement draft Sandy said she sent email to Stig to check the latest changes. Will submit a new version. yang igmp mld snooping draft Updating model based on comments received on the igmp mld model draft. Will do WGLC on the list. yang igmp mld proxy draft Addressed all comments so far. Ian Duncan thinks work should be done. \Stig Venaas pim null register packing draft Update includes the new pim message type extension format for the packed register message types and anycast rp considerations. Dino: i hope the format is exactly like a join/prune message so same code can be used. And snoopers can be used. Stig: it's not. Dino: protocols for ip multicast (pim). why hasn't it been combined with mboned. Mike: decided there was plenty of work in each wg for now. \Stig Venaas reserved bits draft I believe its ready for wglc. Dino: enough types for vendor specific? Stig: would be hesitant because they would reduce the subtypes. Best to just get this draft done. \Stig Venaas pim port pfm draft Not asking for adoption but whether this is useful. Dino: not related to this draft, a bit off topic. are there any requirements to send messages encrypted? Maybe use QUIC. Stig: do have pim port but not sure if anyone has implemented. Toerless: we had that discussion decade back with snooping and security. one compromise to only have authentication and zero encryption so you can see what payload is there. haven't seen it done. zero encryption: nobody needs it according to quic. not sure how difficult it would be to use tls and quick. would love to see it. Dino: do you think the preformance properties of quic merit adding it in? maybe lack of tcp implementaions maybe not. Toerless: I like port. if we adopt anything new that shouldn't be a reason to not use port. so little support for port right now. Stig: also have pim registers support for port. need to gauge if port is the right solution. Toerless: we managed to avoid if you do this across the internet how about the congestion control. answer is port. have a hard time elevating it to the recommended solution. Stig: maybe see if there are better improvements for pim. Mankamana BDR draft Toerless raising whether drafts should have been merged with Sandy's similar draft. Jeffrey also asked if that had been discussed. Stig suggested that drafts reference one another and state what the difference is. Alvaro: clearly state why there are two different drafts for the same problem and why they aren't combined Mike assert packing Stig thinks its a good idea and asking about packing type.... what to announce if you support both. Toerless explaining how people had to deploy in certain ways to avoid asserts. People want LAN for various reasons and asserts need to be handled better. Important to solve this. Jeffrey: Wondering if other ways of avoiding asserts. Can one choose upstream based on what others requested. Toerless says he suggested strong RPF, but people did not like it. Jeffrey: When packing, how long do you wait before you send out Mankamana: Wanted to know how to implement it.... Dino: Says last points critically important. Aggregate means collect subsequent, cannot wait.... Dino: Why asserts happen so often for SM? Asking if DM was a concern. Lenny: Why asserts put in there, explaining why, exchanges.... Toerless: Document about assert problems in exchanges etc in mboned. This is for typical L2 transit LANs. Ice: Supporting Toerless, says better get rid of them than making it better. Dino: Asking if transit LANs could be P2P or not... Toerless: Explaining how to avoid duplicates, strong RPF check would be best. Dino: Strong RPF check would be a tough change. Ice: I wasn't suggesting to stop using asserts. Explaining strong RPF. Dino: Just remove asserts and avoid duplicates for a short amount of time. Toerless: Duplicates causing a lot of problems, strong RPF check would help. Dino: Strong RPF super costly for vendors. Ice: Allowing duplicates can be difficult. Dino: Unicast RPF change affects a lot of flows, prunes would take care a lot at the same time. Lenny: Choices are replace asserts, alternative, don't bother doing asserts. 3rd choice, optimize asserts. Skipping assert maybe makes sense if doing a new implementation Hooman TreeSID SR policy for P2MP Dino: Can one compute reverse paths as well? Hooman, yes Dino: Prefer head-end replication to spray. Spray sounds scary, sounds like broadcast. Jeffrey: There is discussion on the term, agree it could be renamed. Head-end replication is established in a different context. Says some confusion about replication at the root. Toerless IGMP/MLD evolution Toerless says it might make sense to also ask for input on application SSM support. Stig: One challenge is how to reach the right audience. Tim Winters said he could help.