Wednesday (135/150)
start 7/31/2013 9:02 AM          
          WG Status
          co-chairs, 15 min
followed slides
lou re: p2mp framework: thinks doc is ready to go, will come back to chairs.  thinks comments are addressed, authors will reread to make sure.
george: proxy ping, sends thanks to reviewers - will ask for LC.  tictoc draft coming friday, please read.

7/31/2013 9:18 AM
          Loa Andersson, 5 min
loa: points in red are in registry by mistake, need to clean up.
george supports.
stewart: can we leave a tombstone around somewhere?
loa: could do but doesn't cover everything.

7/31/2013 9:22 AM
          Loa Andersson, 5 min
No comments

7/31/2013 9:25 AM
          Carlos Pignataro, 15 min
Wes George: Need people in MPLS WG who work with other WGs to ensure those other WGs are thinking about v6 gap analysis as they touch mpls-related things.
(Ina?): Is there a need for (? this document?) 
Carlos: Not a migration doc, but a what-if.  Also need to make a single view of v6 coverage to understand standards exposure.
Wes: MPLS assumes dual stack but dual stack is a transition mechanism not an end state.  Can't afford to put v4 addrs on infra.  Millions of devices, can't get them all with v4, if they ever need mpls they need v6-only.
George Swallow: this is not a migration doc. 
Loa: poll for wg adoption.
Lou: isn't it already a wg doc?
Carlos: typo

7/31/2013 9:36 AM 
          Mach Chen, 10 min
Greg Mirsky: What impact will SL have on entropy label/ecmp?
Mach: Both can coexist
Greg: Cannot use EL per source if sources identify themselves.  Also, this proposal increases the label stack all the time (just in case we need measurement), cannot toggle SL.  
Loa: is on/off SL function possible?
Greg: No; can propose that loss measurement must be synthetic instead
Lucy: questions on use cases.  
Mach: Clarified use cases.
(Robin? someone?): SL will not impact label merge.  Also, maybe source label is useful for multicast.  
Carlos Pignataro: How do you identify the sources?  How do you distribute unique IDs to each node and understand them network-wide?  
Mach: There are distribution extensions to provide that.  Global label, egress-local-significant label.
George Swallow: for local method w/block+index, how does that work in p2mp since block is allocated by endpoint?
Mach: need to think about.

7/31/2013 9:50 AM          
          Will Liu, 15 min

Sam Aldrin:  Do not want to obsolete LSPID since it is in use elsewhere.
Adrian Farrell: suggest advice from mibdoctors or on how to retire old MIB modules.  also read rfc4990 plase.  
Loa: expect further mailing list discussion

7/31/2013 9:58 AM
          Fan Peng, 10 min

Loa: Doc is already in RT review.  Recommend a MIB expert to review (mib doctors).          

7/31/2013 10:02 AM
          Alia Atlas, 10 min
Lucy: If a router doesn't support MRT, do things still work?
Alia: See 'MRT Island' section
Stewart: Should this be published as Experimental?
Alia: No...architecture is Informational.
Stewart: Question of wide-scale suitability, question for WG
Alia: Doesn't think it needs to be Experimental
Lucy: supports both p2mp and mp2mp?
Alia: MRT works as Mcast FRR alternates.  See architecture doc.  mldp live:live should work too.
Ice: Yes, should work.

7/31/2013 10:11 AM          
          Ice Wijnands, 10 min

Arkaidy, ThompsonReuters: Supports draft.  
Wes: jabber question: does wcast dest mean everyone gets it, or just those on the shared tree?
Ice: Everyone who signals the FEC.
Jeff Tantsura, Ericsson: supports
Loa: number of drafts (WG and LC), requests help to sort out the order in which drafts progress, there are interdependencies.

7/31/2013 10:23 AM          
          Kireeti Kompela, 5min
George: Do split LSPs remerge at the next node?
Kireeti: Yes.
Lucy: What kind of constraints apply to the 80Gb  LSP in the example?
Kireeti: the usual TE constraints.
Lucy: are there constraints on sub-LSP splitting?
Kireeti: Headend, implementation question.
Lucy: Can subLSPs be different size?
Kireeti: up to headend/path computation.
George: all those decisions are local to the headned
Carlos: question on diagram...when sub-LSPs go to different node, you can't merge them,
Kireeti: yes.  FRR is a local action until resignal.
Sam Aldrin: Is FRR for sub-LSPs, or for the whole thing?
George: Think of it as a tunnel interface w/a bundle of LSPs.  they can diverge and reunite as necessary.
Kireeti: can use traditional FRR or rebalancing FRR as needed/possible.

7/31/2013 10:37 AM       
          Yimin Shen, 10 min

George: wants carrier feedback - is this really a problem?  Approach is sound.
Igor Bryskin, Adva: What if P1 can't find a detour but P0 can?
Yimin: Bypass is preestablished.  
Igor: what if it does not exist?
George: crankback to P0
Loa: does this need to go to CCAMP instead? (Debora: MPLS WG seems approprivate)
Ina Minei, Juniper: provider on the draft; some are interested.  
George: need providers to give problem scope one the list

7/31/2013 10:50 AM
          Richard Li, 5 min

No comments.  

7/31/2013 10:54 AM
          Zhenbin Li, 15 min
Jeff Tansura, Ericsson: comment on mldp proxy and static routes.  static routes are exceptions, not normal process.  doesn't see a need for ldp dod proxy.
Ina: proposing protocol enhancements for things that can be done in a local implementation.
Loa: start validating requirements on the list.
George: draft complicated, dense.
Zhenbin: discussed requirements with carriers.
George: perhaps try not to boil the ocean.

7/31/2013 11:18 AM
          Zhenbin Li, 5 min
(discussion, need to transcribe from audio)
Yimin, Eric, George, Jeff, Zhenbin
8/2/2013 9:03 AM

8/2/2013 9:05 AM
 1588 over MPLS
          Karen O'Donoghue, 10 min
George: does not break MPLS but may not work with LDP (BFD packets may not follow the same path as 1588).  also have RSVP feedback (offline)
Ross: what is timeframe for comments since last call has ended?
Karen: get principals together, arrange concall in weeks/months.  
Yaacov via Jabber: don't have to use LDP, there are signaling drafts that go with this draft which don't use LDP.
George: so take LDP part out of this draft
Greg Mirsky: sec. 11 says entropy label must not be used.  requirement in draft may be backwards.
     Some of this work changes the payload without terminating LSP, that breaks MPLS arch.
     Sec. 16 - p2p LSPs are specified but type not provided, need clarification
Yaacov via Jabber: re previous comments, do not use LDP.  Agrees with the comment on entropy label, but it has other problems.  layer violation is inherent to transparent clocking.
Greg: layer violation not inherent in transparent clocking if you terminate the LSP
Loa: send draft to Review Team to look for issues
Karen: needs real-time conversation, email/meetings have been going on for a while
Stewart Bryant: most people may not understand the properties of the new type of LSP this draft creates.  need succinct description of new FEC that is being created and the properties it needs.
Karen: wants to move draft forward, ok to work on technical solutions but need to find out if this solution breaks mpls or precludes other solutions
Stewart: benign but not general.  needs succinct definition of new proposed FEC type and its properties.
Sharam Davari: proposing a TE LSP for timing only where midpoints are aware of the LSP's properties.  Can address comments, but the main question is whether they can go forward.
George: Out of time.

          Eric O., 10 min

Robin: Not that many colors needed. 
Adrian: Computed or hoo-by-hop.
<< need to transcribe from audio >>
end at 8/2/2013 9:37 AM

8/2/2013 9:38 AM
    updates to psc from chairs (1 slide)

8/2/2013 9:39 AM
          Alessandro D'Alessandro
George: IETF process allows for group progression but only in editor's queue; rest of process has no lockstep.  If we need lockstep we can have 1 doc instead of 4.
Ross: could also best-effort attempt to keep things together.  
Loa: we can make it work either wya
Ross: if all docs are adopted, will we end up with one PSC method that satisfies IETF and ITU-T?
Alessandro: Yes.  One protocol.
Ross: Maybe 'unified mode' since it makes everyone happy.
Sharam Davari: goal is to unify methods.  but do we also want to change packet format?
Alessandro: will not modify the packet format when porting to G.8131.
Sharam: Will not have identical format?
Alessandro: will have the same packet format, same semantics, use TLV to manage behaviors

          Hui Deng, 10 min
8/2/2013 10:01 AM
Greg Mirsky: should be two docs...interconnecting ring protection and shared ring protection.  interconnected ring protection is an instantiation of existing RFC, need to reuse that as much as possible.  for shared, there is a discussion on shared mesh, have the authors looked at them?  Will send a pointer.
Sharam: draft is sufficiently different from 6974 but this draft should only have what it adds, not repeat what's already documented.  useful draft but should reuse not repeat.

8/2/2013 10:14 AM
          Santiago Alvarez, 10 min
Ina Minei: draft restricts this to DU mode. why?
Santiago: Draft references both DU and DOD
Ina: Not clear in draft.
Ina: why is the egress the one who advertises color interest?  makes more sense for headend to do it.  egress may not know all the types of traffic that terminate on it.  need clarity around what you do if you're a midpoint and you don't have a path for that color.  seems underspecified. 
Santiago; two modes.  one in which midpoint advertises multiple labels if they exist.  second is the midpoint only advertises if the egress indicates desire.  no headend mechanism defined.
Ina: needs more text around how this would work.
Ina: draft relies on LDP over RSVP, but is this a generic mechanism (e.g. LDP over physical)?  if a certain color doesn't exist end-to-end, what do I do?
Santiago: take questions to list
Sam Aldrin: what is the impact on OAM?
Santiago: draft defines OAM extensions to test per color.
Ross: take questions to list.

          Zhenlong Cui, 10 min
8/2/2013 10:23 AM
  No comments.
Ross: need more comments from WG.

          Richard Li, 10 min
8/2/2013 10:31 AM
Sharam: "requires 17 tables" is not correct...hash or tcam implementations work differently.  
Richard: performance is unpredictable, other reasons
Sharam: will explain offline
George: read rfc3107?  label stacks are already permited in bgp.
Richard: implementations only carry one
George: stacking is permitted, just not used
Zhenbin Li: (didn't get comment)
Stephane Litkowski: is use case realistic?  How many interfaces on a PE?  Don't have 1M interfaces on one box.
Richard: One or two boxes as L3VPN/NVE edge.  Open to collaboration.  Some operators prefer mapping vnet ID to VPN label. 
George: Proposal (other wg?) for vlan ID rather than RT.  Many operators didn't want.  
Ross: does one box really have 16M labels off of it?  Needs more thought.
Richard: Agree, needs more discussion.

          Zhenbin Li, 10 min
(forgot to get timestamp)
Eric Osborne: isn't this just context labels:
Robin (Zhenbin): optimized for a particular implementation method
George: how do I know on the wire whether it's a 'base label' or a regular label?
Robin: context labels are not in use today.
Sharam: this is just context labels no matter the implementation.
George: is there a provide who carries Internet routes in LDP (slide 2 case 1)?  Also, L3VPN already has stacking mechanisms.  On 3rd case, is a universal vlan ID useful?  
Stephane Litkowski: is this really necessary, and if it is can we use context labels?
Ross: time up.  also, MRT authors should look at case 1.