MSDP 12/11/00 Meeting Minutes David Meyer Chair (dmm@cisco.com) Sean McCreary Scribe (mccreary@colorado.edu) Maddogs update from Dave Meyer Meeting consensus from 12/10 is that BGMP for IPv4 isn't going to happen Bill Fenner: It won't be implemented unless current solution fails to meet demand Dave: No way to collect data on how much SSM is in use Must be on path from sender to receiver to see any traffic MSDP isn't needed to establish SSM distribution trees So MSDP doesn't need to scale up as much as for ISM Near-term solution for IPv4 inter-domain multicast is MSDP and SSM Dave asked for anyone to object to this statement, no one did He declared rough consensus on this point MSDP was originally implemented to provide anycast-RP for PM-SM It has now been adopted as a longer-term interdomain multicast routing protocol BGMP will not be required in the forseeable future. Dave Thaler asked about MSDP implementation for IPv6 Dave Meyer: No current work in progress Maybe should find another way to do RP redundancy in IPv6 Dave Thaler said the protocol has nothing that would prevent an IPv6 implementation Rough consensus: MSDP for IPv6 is dead Tom Pusateri: Can the PIM working group decide to use MSDP for IPv6? Dave Meyer: If they decide to do this, the MSDP WG will work on it Question: Operational overhead of running BGMP and MSDP in parallel is prohibitive Don't want different inter-domain protocol for IPv4 and IPv6 Tom Pusateri: IETF role is to come up with the best solution Customers can ask vendors for specific solutions outside of the IETF process MSDP draft disposition (02 vs. 06) Dave asked for current implementations of MSDP Cisco Juniper Lucent NEC BT Current deployment is 02, 06 is current draft status 02 should move to RFC w/historic classification 06 should move to RFC w/experimental classification Bill Fenner would like sanity-check of peer-RPF rules he added before the drafts are moved to RFCs Problem: No one is implementing 06, and at IETF 48 decision was made to ignore differences between 02 draft and current implementation as MSDP was `temporary' Tom: Need to document differences between 02 and currently deployed implementations, especially hold-down Dave: 02 captures what is deployed better than any other draft If we make a new rev, would it include peer-RPF? Bill Fenner: The rules in 02 are how MSDP is implemented in Cisco IOS However, these rules allow loops Loops are unbreakable and difficult to find in current implementations Tom: With holddown, the amount of control traffic generated by loops is low enough that it doesn't present an operational problem. Question: Why not go to standards track if MSDP is forever? Question: We need a draft describing current MSDP implementations so we can understand exactly what is under discussion Bill Fenner: 06 would be good to use as a base for this discussion Dave called for consensus on moving 06 to experimental RFC Audience approved Dave called for consensus on moving 02 to historic RFC Audience approved Dave noted that a previous snag to moving MSDP to standards track was that GRE was published as an informational RFC This has changed (GRE is undergoing standardization) Bill Fenner gave an update on the MSDP MIB He has an unpublished version with changes to support IPv6 addresses Current efforts are underway to implement currently published version w/only IPv4 address support Perhaps the changes to support IPv6 are unnecessary? Question: Using INET ENDPOINT identifiers won't hurt anything Will make MIB more general IESG is adamant about all new MIBs use INET ENDPOINT identifiers No implementers were present in the room Bill asked if the new MIB should be published with IPv6 support Audience approved 02 -> 06 Transition document Needs to reflect changed status of MSDP, to permanent inter-domain multicast routing protocol for IPv4 MSDP traceroute Bill: It would be really nice to have, not worth blocking progress to complete Rob: Why not use SNMP to gather data from routers along distribution tree rather than writing a new protocol Dave Thaler: SNMP is only meant for data gathering within a single administrative domain Bill: We want arbitrary users to be able to perform MSDP traceroutes Don't want access-control model of SNMP SNMP would only be used to initiate traceroute, not for queries between routers Other than initiation protocol, there isn't that much left to do on the MSDP traceroute spec Dave: MSDP debugging currently requires manual analysis of SA state hop-by-hop A traceroute protocol would be very helpful MSDP-specific forwarding extensions (MSDP-FE) presentation by Masahiro Jibiki NEC's implementation of MSDP is based on 06 draft He requested that new packet types be added to the 06 draft and that the draft be accepted as a working group document Dave deferred both points to discussion on the mailing list