Tuesday, 15 November 2011< ^ >
Room Configuration

[00:51:34] Yuri Nawata joins the room
[00:58:03] jlcJohn joins the room
[00:59:02] Bill joins the room
[00:59:28] Shane Amante joins the room
[00:59:42] <Shane Amante> howdy folks, i'll be the jabber scribe for today
[00:59:58] <Shane Amante> let me know any questions you want me to relay to the mic
[01:00:34] iLya Varlashkin joins the room
[01:05:22] <Shane Amante> WG meeting starting
[01:05:24] <Shane Amante> note well
[01:05:51] <Shane Amante> agenda bashing
[01:06:02] Stewart Bryant joins the room
[01:06:19] <Shane Amante> WG status
[01:06:28] <Shane Amante> 3 new WG docs
[01:07:10] <Shane Amante> deprecate-as-sets - delayed pub. due to concerns about change to BCP
[01:07:32] <Shane Amante> danny: was going to be BCP, but does not say "Updates: 4271"
[01:08:25] <Shane Amante> 2 step process: publish deprecate-as-sets now, then go back later and update 4271 later
[01:09:12] TACHIBANA toshio joins the room
[01:09:16] <Shane Amante> stewart: if really concerned, then do 1-page stds track doc to say Updates 4271
[01:09:59] <Shane Amante> everyone agrees 2-step process is OK
[01:10:32] <Shane Amante> 3 new WG docs: wkumari-idr-as0, gr-notification & idr-enhanced-gr
[01:11:34] <Shane Amante> Graceful Restart
[01:12:51] <Shane Amante> GR restart re-spin has begun
[01:12:57] <Shane Amante> Enke holds editing pen, at the moment
[01:13:47] <Shane Amante> need to clarify operational modes, language overall, etc.
[01:14:08] <Shane Amante> now is a good time for reqmt's input
[01:14:45] <Shane Amante> Route Servers
[01:15:00] <Shane Amante> draft-jasinska-ix-bgp-route-server is now @ -03
[01:15:10] <Shane Amante> authors requesting WG adoption
[01:15:22] <Shane Amante> will hold WG call for adoption soon
[01:15:57] <Shane Amante> SueH to now talk about error handling
[01:16:10] <Shane Amante> Slides: "Upcoming BGP Changes for Error Handling"
[01:16:17] <Shane Amante> Slide: Types of Errors
[01:17:24] nco joins the room
[01:17:34] <Shane Amante> Q: poor service or no service? in past, err'ed on side of no service
[01:17:48] <Shane Amante> new reqmt's from GROW are "poor service" vs. no service
[01:17:57] <Shane Amante> Slide: Problems iwth Poor instead of "No Service"
[01:18:14] <Shane Amante> don't want errors to hang-on forever
[01:19:09] <Shane Amante> Slide: Errors with "Treat as Withdraw"
[01:19:49] <Shane Amante> Slide: How to Recovering RIB's
[01:21:14] <Shane Amante> Slide: Call to Operators
[01:21:37] <Shane Amante> Final call for operators for input on reqmt's for Error/Monitoring, but very concerned about complexity wrt "solutions"
[01:21:57] <Shane Amante> Questions?
[01:22:59] <Shane Amante> Rudegir: In Sept. Attr. 128 lead to major problems
[01:25:17] <Shane Amante> Robert: How do you envision BMP being part of this?
[01:25:57] <Shane Amante> SueH: anything thing that GROW suggests as a result of input from operators, IDR wants to consume that as input to this solution space
[01:26:26] <Shane Amante> ChrisM: Wrt BMP, if a session is going to die, send the UPDATE out to a monitoring station before dying
[01:26:47] <Shane Amante> Robert: later revs of BMP removed capability to replay errors
[01:27:33] <Shane Amante> ChrisM: overall point is to figure out when to log, how to log, etc.
[01:28:09] <Shane Amante> 1st preso is going to be Enke on "Error Handling"
[01:28:36] <Shane Amante> Revised Error Handling for BGP Update Messages, by Enke
[01:28:48] <Shane Amante> draft-ietf-idr-optional-transitive
[01:29:29] <Shane Amante> Slide: Reasoning
[01:30:44] <Shane Amante> Lesson learning: error handling must be at the route level, not session level
[01:31:32] <Shane Amante> Slide: Summary of Changes
[01:33:33] <Shane Amante> Slide: What about iBGP?
[01:34:04] <Shane Amante> - treat as withdraw may not be enough for iBGP
[01:36:01] <Shane Amante> - not pursuing automation of procedures in protocol, b/c issues in eBGP typically dominate and it would require significant changes in protocol
[01:37:01] <Shane Amante> Slide: Errors with Attribute Flags
[01:37:24] <Shane Amante> - can be fixed locally; no need to reset session
[01:37:59] Danny McPherson joins the room
[01:38:23] <Shane Amante> - another option would be treat-as-withdraw
[01:38:28] <Shane Amante> Questions?
[01:38:40] <Shane Amante> Lots of folks have read the current version of the draft
[01:38:48] <Shane Amante> Next preso
[01:38:59] <Shane Amante> oops, one comment from Alton Lo @ Cisco
[01:39:22] <Shane Amante> Alton: lots of problems because of Optional, Transitive Attr.
[01:40:14] <Shane Amante> Q: should we consider keeping optional, transitive Attr around?
[01:41:56] <Shane Amante> Danny: optional transitive is a problem b/c it tunnels information through, but affects adj. with others
[01:42:56] <Shane Amante> Next preso: "Automatic Route Target Filtering for Legacy PE's" by Alton Lo
[01:43:16] <Shane Amante> draft-ietf-l3vpn-legacy-rtc-00
[01:43:24] <Shane Amante> Slide: "Motivation"
[01:45:18] <Shane Amante> Slide: "Legacy PE Method"
[01:45:57] <Shane Amante> Slide: "Route-Filter routes handling by RR"
[01:46:45] <Shane Amante> Slide: "Exchanging RTC between legacy PE and RR"
[01:47:18] <Shane Amante> Slide: "Status"
[01:47:54] <Shane Amante> Slide: "Next Step"
[01:48:00] <Shane Amante> Questions?
[01:48:29] <Shane Amante> John: start discussion on list to poll for WG adoption? Decent # of folks in room read this draft.
[01:50:00] <Shane Amante> Danny: when moving from RFC 1996 to RFC 2796, we removed ability on RR to reflect back route to the client that originated the route. This draft is allowing routes to get reflected back to client that originated them, based on RTC automatically cfg'd on RR toward client.
[01:51:39] <Shane Amante> Robert: have operational concern. if some RR's support this & some RR's don't, then need to re-home PE's to RR's that do support this feature.
[01:52:37] <Shane Amante> Keyur: need to run this on all RR's in a domain.
[01:52:58] <Shane Amante> SueH: what happens if you think you've rolled this out to all RR's, but you haven't? (How does this draft handle it).
[01:54:57] <Shane Amante> Danny: some implementations of RR's today don't reflect a route back to the client that originated the route to the RR. If this isn't understood, then it could result in blackholing of that route, which is similar to what is being proposed here.
[01:55:06] <Shane Amante> Will take these questions to the list.
[01:55:35] <Shane Amante> Next preso: "BGP Persistence" by Bruno
[01:55:41] <Shane Amante> draft-uttaro-idr-bgp-persistence
[01:55:52] <Shane Amante> Slide: "Why?"
[01:57:02] <Shane Amante> Slide; "What? (1): BGP session failure"
[01:58:04] <Shane Amante> Slide: "What? (2): BGP session re-establishment"
[01:58:40] <Shane Amante> Slide: "Example of use cases"
[01:59:40] <Shane Amante> Slide: "Caveats"
[02:01:35] <Shane Amante> Slide: "Why not Grace Restart (GR)"?
[02:02:59] <Shane Amante> Slide: "Next Steps"
[02:03:45] <Shane Amante> Questions?
[02:04:18] <Shane Amante> Enke Chen: this draft is basically GR with a few tweaks
[02:04:40] <Shane Amante> Enke: since GR re-spin is going on, we should try to accommodate this into GR re-spin
[02:06:03] <Shane Amante> Enke: no diff. between GR and this draft in terms of fwd'ing path being available, when BGP session goes down it's the same thing
[02:06:17] <Shane Amante> Enke: can introduce concept of min. stale timer in GR
[02:07:00] <Shane Amante> Bruno: if we want to revisit what GR spec says, I'm OK with that
[02:08:12] <Shane Amante> Enke: GR spec says to only affect fwd'ing plane locally to not disturb traffic fwd'ing
[02:08:43] <Shane Amante> Enke: if session is not re-established after 68 mins, then stale path is purged
[02:09:06] <Shane Amante> Enke: if want to maintain routes for longer that's based on receipt of EoR
[02:09:27] <Shane Amante> John: Enke is suggested that this draft re-spun as reqmt's for GR re-spin
[02:09:47] <Shane Amante> John: that seems to be agreement between Bruno & Enke
[02:10:47] <Shane Amante> Keyur: regardless of marking routes persistent or not, when session goes down, then no new routes are going to get through
[02:11:40] <Shane Amante> Keyur: all the marking does is to help nodes downstream to know that new updates to route will get through
[02:12:29] <Shane Amante> Robert: 90% of reqmt's in this draft can be addressed by GR
[02:12:46] <Shane Amante> … addressed by GR re-spin
[02:13:23] <Shane Amante> Robert: 10% is not addressed by GR. marking is needed to influence path selection externally to domain, (inside domain can use cost community).
[02:14:14] <Shane Amante> JohnS: don't really have an answer to that question, yet, but should use existing tools where possible
[02:16:15] <Shane Amante> Ilya: re: Best Path Selection … what happens if you have 2 x RR's and session to one of them dies and then primary AC also dies and supposed to go over backup. problem is stale path is still preferred through primary AC which is down, how do you handle that?
[02:18:17] <Shane Amante> JohnS: for stale routes, they should be treated as least preferred, therefore that means any backup route would get preferred/used
[02:18:48] <Shane Amante> Next Preso: "MTU Ext. Community for BGP-4" by Jie Dong
[02:18:54] <Shane Amante> Slide: "Background"
[02:19:10] <Shane Amante> - MTU discovery for Inter-AS LSP for Inter-AS Option C, CsC, etc.
[02:19:19] <Shane Amante> Slide: "Solution Review"
[02:19:49] <Shane Amante> Slide; "Changes in -01"
[02:20:42] <Shane Amante> Slide: "Next Steps"
[02:20:59] <Shane Amante> Questions?
[02:21:11] <Shane Amante> SueH: will take WG adoption question to the list
[02:21:48] <Shane Amante> Next Preso: "Northbound Distr. of Link State and TE Info. using BGP" by Hannes
[02:22:06] <Shane Amante> Hannes apologies latest copy of slides aren't on Web, yet
[02:22:32] <Shane Amante> draft-ietf-idr-ls-distribution-00 (was: draft-gredler-bgp-te-01)
[02:23:41] <Shane Amante> - wanted to clarify that this draft is trying to distribute info. for applications like ALTO, PCE, etc. … not to do LS computations inside BGP
[02:23:45] <Shane Amante> LS = Link State
[02:23:50] <Shane Amante> Slide: "Motivation"
[02:24:01] <Shane Amante> Slide: "Use Case - Alto Serviers: Multi-Area IGP Topology"
[02:24:23] <Shane Amante> - ALTO server need to know all areas in topology
[02:24:48] <Shane Amante> - manual crafting of "IGP peering" topology is tedious & error-prone
[02:25:05] <Shane Amante> Slide: "Use Case: Path Computation Element"
[02:26:41] <Shane Amante> - No clash with work going on in PCE, b/c PCE Architecture is fairly wide-open. Prevailing model is a distributed signaling model. PCE hasn't specified how to sync. TEDB's across AS'es.
[02:26:53] <Shane Amante> Slide: "Major Changes since IETF81"
[02:28:04] <Shane Amante> - support for Node Attr. (O/L bit, Capabilities), Multiple Protocols sharing a link (OSPF/ISIS, L1/L2 Intra-POP links), support for MT/MI extensions; Added support for OSPF/IS-IS Area ID
[02:28:29] <Shane Amante> Slide: "Transcoding TE Link Info info BGP NLRI"
[02:28:57] <Shane Amante> - 2 NLRI types carried in MP_REACH and MP_UNREACH_NLRI
[02:29:03] <Danny McPherson> draft-ietf-sidr-bgpsec-ops-01.txt: The author wishes to thank the BGPsec design team.
[02:29:31] <Shane Amante> Slide: "Transcoding TE Link Info into BGP NLRI TED (SAFI 1)"
[02:29:51] <Shane Amante> Slide: "TED NLRI TLV Types"
[02:30:25] <Shane Amante> - NLRI advertises a 'unidirectional link' in Link Descriptor TLV's
[02:31:26] <Shane Amante> Slide: "Transcoding TE Link Info into BGP NLRI TED (SAFI 128)"
[02:32:05] <Shane Amante> - SAFI 128 identical to SAFI 1, but it has a RD "flavor"
[02:32:15] <Shane Amante> Skipped a couple of slides, now on "Node Attributes"
[02:32:37] <Shane Amante> Slide: "Next Steps"
[02:32:50] <Shane Amante> - would like feedback; accept as WG item?
[02:32:58] <Shane Amante> SueH: has this been reviewed by PCE WG?
[02:33:01] <Shane Amante> Hannes: Yes
[02:33:10] <Shane Amante> Questions?
[02:34:08] <Shane Amante> PCE co-chair: I believe there is no overlap; however, that's not necessarily the consensus of the PCE WG.
[02:34:27] <Shane Amante> John/Sue: can you poll the PCE WG to make sure there is no overlap so this can proceed in IDR?
[02:34:47] <Shane Amante> Hannes: use this as mechanism to synchonize TED DB's
[02:35:11] <Shane Amante> Hannes: PCE is really the receiver of this information
[02:36:34] <Shane Amante> ???: this overlaps with distribution of TE info. across IGP Areas. In IGP, TE protocols we already have this in Type-9, 10, 11 LSA's in OSPF. If we use Type-11, we can already distribute this from one area to another area in one AS.
[02:37:03] <Shane Amante> Hannes: for OSPF, this is correct; however, majority of LSDB only has area or level-scope
[02:37:31] <Shane Amante> Hannes: other major thing to address is "Seamless MPLS" where we have disjoint IGP's
[02:37:51] <Shane Amante> Hannes: this proposal can do Inter-Area and Inter-AS, etc.
[02:38:18] <Shane Amante> Wei: for Areas & AS, some SP's don't want to distribute some info. about whole topology outside of their AS.
[02:38:42] <Shane Amante> Hannes: all these extensions are optional, so there's no need to disclose all this info. externally
[02:39:07] <Shane Amante> Hannes: in the draft, there is a discussion about "aggregation" to adv. a set of physical links as a single Virtual Link
[02:39:45] <Shane Amante> Ilya: mentioned Inter-AS Option C in slides, but nothing in draft — can use it for Inter-AS Option B.
[02:39:59] <Shane Amante> Hannes: can use this for Inter-AS Option B, as well
[02:40:29] <Shane Amante> Ed Crabbe: this is _a_ state distribution mechanism, not _the_ state distribution mechanism … so others can come along with something else later
[02:40:59] <Shane Amante> Ed: also, want to emphasize what Hannes said that IGP's are disjoint, so this is needed to share information northbound to PCE's
[02:41:22] <Shane Amante> SueH: if you don't want to distribute the whole DB does this still work?
[02:42:02] <Shane Amante> Hannes: as long as you advertising a continuous link to PCE's, this should work
[02:43:52] <Shane Amante> Wei: one concern is about scalability, if we add more information such as TE, then how does this affect that?
[02:44:06] <Shane Amante> Hannes: largest SP's in the world have about 6 − 8K links ...
[02:44:34] <Shane Amante> Ed Crabbe: this is intended for targetted dstributed of state, not full scale distribution of state amongst all routes in an AS
[02:46:38] <Shane Amante> Hannes: it's only a few thousand entries for even the largest networks, so each BGP speaker needs to hold the whole thing, but shouldn't be substantially more load
[02:47:02] <Shane Amante> Rudegir: does the same BGP instance (magic carpet) need to hold/maintain this info?
[02:47:23] <Shane Amante> SueH: need the draft to discuss scaling characteristics in a future version of the draft
[02:47:49] <Shane Amante> Ilya: could use multi-session to separate out this AFI from something else
[02:48:10] <Shane Amante> Hannes: knowing Rudegir, expect that this would be in a separate routing process altogether
[02:48:40] Danny McPherson leaves the room
[02:49:19] <Shane Amante> Next Preso: "Application of draft-varlashkin ...", by Ilya
[02:49:35] <Shane Amante> Slide: "Overview"
[02:50:13] <Shane Amante> - next-hop reachability is not transitive … next-hop SAFI addresses this issue
[02:51:07] <Shane Amante> - if RR can reach next-hop, that doesn't mean RR client can reach NH (or, client has diff. IGP metric to NH)
[02:51:43] <Shane Amante> Slide: "The Issue Illustrated"
[02:53:16] <Shane Amante> Slide: "BGP Exchange Example"
[02:54:30] <Shane Amante> Slide: "Solution Illustrated"
[02:55:01] <Shane Amante> Slide: "Last Slide"
[02:55:17] <Shane Amante> Questions?
[02:55:48] <Shane Amante> JohnS: if you take Optimal RR and add on this draft, then you can get "for free" NH reachability
[02:55:54] <Shane Amante> Ilya: Yes
[02:56:04] <Shane Amante> JohnS: yeah, it allows you to fulfill reqmt's in 3rd draft
[02:56:25] <Shane Amante> JohnS: this presents a use case for why to adopt the other drafts
[02:56:35] <Shane Amante> Lou: can you accomplish the same thing with add-path?
[02:57:32] <Shane Amante> Ilya: add-paths doesn't distribute info. based on vantage point of RR-client
[02:58:16] <Shane Amante> Lou: will RR ask for all next best-hops from all clients?
[02:58:31] <Shane Amante> Ilya: not doing this each time for every prefix. just doing this once.
[02:58:53] <Shane Amante> Lou: last observation is trade-off of adding complexity to RR vs. pushing for info. over BGP
[02:59:39] <Shane Amante> Lou: 2 mechanisms that can accomplish similar things, do we want both?
[03:00:18] mjbarnes joins the room
[03:00:23] <Shane Amante> JohnS: reiterate that a lot of this discussion previously happened when we adopted the NH-cost draft … so, revisit that discussion
[03:00:39] <Shane Amante> Robert: this draft is not competing with add-paths
[03:01:25] <Shane Amante> GeorgeS: clarification: it's not very chatty, but don't you have to do this each time the IGP changes
[03:01:34] <Shane Amante> Ilya: Yes
[03:02:23] <Shane Amante> Ilya: Even in fairly large network, there will only be a few hundred NH's … so, that's quite limited size of SAFI
[03:03:05] <Shane Amante> Jakob: router is mostly busy when it first starts up, not so much at other times … so, this will delay initial convergence
[03:03:17] <Shane Amante> Ilya: can initially use IGP metrics first time
[03:03:46] <Shane Amante> Jakob: router may have labelled path to NH and iP path to NH, how does router know which NH to use?
[03:04:40] <Shane Amante> Ilya: should take the question to the list when you have mixed environment. This draft was mainly motivated b/c P routers have no way of fwd'ing packets b/c they don't have a full RIB/FIB
[03:05:24] <Shane Amante> Robert: To address George's comment on chattiness … we can still apply thresholds to not advertise new NH's when you're too busy
[03:06:01] <Shane Amante> SueH: we are doing things in lumps, e.g.: Error Handling, Path Selection, etc., so please be patient
[03:06:30] <Shane Amante> That's last topic.
[03:06:37] <Shane Amante> WG meeting concluded.
[03:06:40] iLya Varlashkin leaves the room: I'm happy Miranda IM user. Get it at
[03:07:30] Shane Amante leaves the room
[03:09:51] TACHIBANA toshio leaves the room
[03:14:45] Yuri Nawata leaves the room
[03:16:31] mjbarnes leaves the room
[03:18:25] Bill leaves the room: Computer went to sleep
[03:21:45] Stewart Bryant leaves the room
[03:26:24] jayb joins the room
[03:26:45] jayb leaves the room
[03:28:00] Bill joins the room
[03:30:36] nco leaves the room
[03:37:03] Bill leaves the room
[03:50:10] Shane Amante joins the room
[03:56:05] Shane Amante leaves the room
[04:05:03] jlcJohn leaves the room
[05:01:56] nco joins the room
[05:04:07] Stewart Bryant joins the room
[05:10:09] nco leaves the room
[06:27:16] Stewart Bryant leaves the room
[06:27:43] Stewart Bryant joins the room
[07:06:14] Stewart Bryant leaves the room
[07:19:09] Stewart Bryant joins the room
[08:28:25] Stewart Bryant leaves the room
[09:20:02] Stewart Bryant joins the room
[09:35:18] Stewart Bryant leaves the room
[15:22:26] Stewart Bryant joins the room
[17:26:23] Stewart Bryant leaves the room