[00:17:20] --- ignas has joined
[00:32:40] --- ggm has joined
[00:32:45] <ggm> working agenda
[00:32:49] <ggm> FSM issues
[00:34:21] --- avri has joined
[00:34:50] <ggm> docs without FSM issues. MP BGP/2858 revisions - extended communities - Cease sub-codes
[00:35:05] <ggm> AS confed revisions, 4 byte AS ID
[00:36:04] <ggm> call for implementation reports on 4 byte AS ID
[00:36:20] <ggm> AS-wide ID. still needs to go into last call.
[00:37:58] <ggm> docs with FSM issues
[00:38:41] <ggm> Drafts: graceful restart, route refresh, Dynamic capabilities. Authors can contact WG chair to get info on FSM additions Can do FSM changes in parallel or as separate drafts
[00:40:16] <ggm> Docs depending on docs
[00:40:30] <ggm> MPBGP, ORFs depend on Route refresh,
[00:41:37] <ggm> implementations solicited, need help to progress work
[00:41:59] <ggm> BGP peer restart backoff
[00:42:29] <ggm> (susan hares)
[00:43:09] <ggm> private draft. separated from base draft. whats the problem? peer auto-restart causes peer to go up/down rapidly. shakes the infrastructure. sent, delete routes, ping, back up, repeat.
[00:44:41] <ggm> engines in FSM to do exponential/step/no-hold timer stuff.
[00:44:55] <ggm> should this be standardized?
[00:45:18] <ggm> feedback solicited. accept as WG draft, take to ML.
[00:46:11] <ggm> zinin: small comments. draft says, informational RFC. should say that. target status. don't say its RFC. make sure people not confused. second thing. looking at substates. wondered if we could do same functionality with existing states.
[00:46:16] <ggm> Susan: no. emphatically.
[00:46:29] <ggm> Susan: other comments?
[00:46:54] <ggm> no comments. taken to list
[00:48:05] <ggm> Deternministic Route redist into BGP Enke/Jenny
[00:49:11] <ggm> how RIB determines path. uses 'administrative distance' BGP has LOCAL_PREF,AS_PATH,MED,IGP.
[00:49:13] <ggm> how to reconcile
[00:49:30] <ggm> route redist into BGP is common.
[00:50:24] <ggm> example of problem on slides.
[00:51:21] --- Bill has joined
[00:52:53] <ggm> problem in instance is with two BGP paths, one from R1, via IBGP, one locally redist, with default setting, order of arrival time of paths impacts routing behaviour.
[00:53:21] <ggm> in example, all 3 routers converge on the primary path
[00:54:20] <ggm> when local config path arrives first, local path is seelected due to IGP metric. then adv. to IBGP peers, r2/r3 can wind up using the
[00:54:45] <ggm> backup path. extra route specific BGP can be added, to elimiate non-deterministic behaviour but has operational cost
[00:58:26] <ggm> benefits. eliminate non-deterministic routing. avoid BGP config load, and be computationally efficient. draft is -chen-bgp-redist-00.txt
[00:58:49] <ggm> future revs suggested by curtis: expand on admin-distance. re-title" use of admin distance in BGP route selection"
[00:58:57] <ggm> [end]
[00:58:59] <ggm> Q
[00:59:11] <ggm> Andrew Lang. alcatel. with service provide hat on.
[00:59:40] <ggm> when deploy, use localpref. havn't found it expensive admin., but I dont find that. just normal config template. works. concenred about changing route selection process for something with trivial workaround
[01:00:08] <ggm> Enke: completely backward compatible
[01:00:40] <ggm> takes time to work out: why happening. whats going on. old problem.
[01:01:44] <ggm> enke. already configured, fine. doesn't break.
[01:01:50] <ggm> Andrew: not neccessary. education issue.
[01:01:57] <ggm> Enke how to reach? thats the thing.
[01:02:10] <ggm> Andrew, same as for any other current BGP problem
[01:02:49] <ggm> Barry friedman with cisco.
[01:03:15] <ggm> using admin distance lacks some generality. eg. may have couple of different IGPs. may want to prefer one, by lowering or raise to
[01:03:34] <ggm> prefer remote exit point, would have conflict. probably not appropriate to overload admin distance for use in BGP.
[01:03:45] <ggm> Enke can make high or low. thats fine.
[01:04:36] <ggm> Barry: admin distance is for RIB. may work in RIB, but not in BGP. cust has to make it work in both. may be better to separate and get more generality
[01:06:10] <ggm> <?> nice problem to solve, existed a long time. is point/narrow solution. if expand problem space a little bit, eg 2 cust sites with link
[01:06:25] <ggm> between, now, this solution needs 4 route based configs
[01:07:02] <ggm> granularity required changes. so now need to overload different ways.
[01:07:26] <ggm> ENke is that on the same box? is one local and one learned? if learned through iBGP have problem. I showed using statics but happens using BGP as well.
[01:09:04] <ggm> Enke one way or the other, have to config. you have this problem. in that sense, its a general problem. taking admin distance into consideration solves that problem. talk offline.
[01:10:14] <ggm> Zinin trying to analyze the problem. with a specific impl, 2 routes can be selected in 2 places. BGP and RIB. what your're trying to do, is to make them consistent inside certain time constraints. my Q is, how reasonable is it to assume this happens, in two places, in all impl. is it impl-specific? I think I can see others with only one place. trying to figure out if applicable
[01:10:25] <ggm> Enke ?
[01:10:44] <ggm> Zinnin. not saying HAS to happen in 2 places. I am saying there ARE impl where it happens in 1 place. BGP selection at same time
[01:10:49] <ggm> as admin distance comparison
[01:10:57] <ggm> Enke. current deployment pattern
[01:11:10] <ggm> there for all impl.
[01:11:26] <ggm> Zinin agree its there, but only non-deterministic in cases with 2 places.
[01:12:00] <ggm> will take offline. second Q. new step in route selection process, may change traffic pattern for existing path selection?
[01:13:11] <ggm> ENke no, don't think so.
[01:13:27] <ggm> <?2? does change behaviour. by definition not deterministc before, deterministic after
[01:13:33] <ggm> zinin. ok. have to check every config.
[01:13:54] <ggm> <?2> not saying do this, but am saying this changes things.
[01:14:19] <ggm> yacob: continue zinin view. only have 1 part. in that case, we are dealing with impl-matter issue.
[01:14:22] <ggm> zinin thats the Q.
[01:14:32] <ggm> yacob. if in fact impl matter, document and leave it at that.
[01:14:39] <ggm> zinin authors meant it to be?
[01:14:45] <ggm> ENke? either is fine. docu or standard
[01:14:50] <ggm> sorry inf or standard
[01:15:04] <ggm> <?2> if changing path selection, can't make it informational.
[01:15:09] <ggm> yacob purely local matter
[01:15:13] <ggm> zinin path selection is not local matter
[01:15:29] <ggm> yacob: need to answer q.
[01:15:56] <ggm> roul may be impl matter. one assumption is route selection in BGP, rib gets route. if in one place, local matter
[01:16:32] <ggm> <?1> irrespective of how, is not really local thing. has to be consistent wth other router. then desired outcome will not be attained.
[01:17:00] <ggm> would require edge router upgrade
[01:17:14] <ggm> enke why? last point. is enough to upgrade one node.
[01:17:35] <ggm> <?1> depending on determinisim of two conflicting impl, either side could make choice differently.
[01:17:37] <ggm> ENke no
[01:20:08] <ggm> Ronald/MCI. discuss how this works. using localpref today. roll out code deploys this change, now admin weight takes priority. will existing config behaviour change?
[01:21:02] <ggm> Enke depends.
[01:21:27] <Bill> <?1> was Gargi Nalawade
[01:22:33] <ggm> Gengis: <missed it. sorry.>
[01:23:11] <ggm> Zinin: small idea. seems problem can be solved without changing path selection process. mapping. decision can be made both ways by correlation
[01:23:21] <ggm> Yacob: take to list.
[01:23:24] <ggm> [end]
[01:24:03] <ggm> BGP based auto discovery for L2VPNs
[01:24:22] <ggm> also being presented in l2vpn grp. looking to alex to say where it does for a home.
[01:24:44] <ggm> tying to reduce configuration. presenting BGP mechanisms to get feedback.
[01:24:52] <ggm> (susan)
[01:25:27] <ggm> uses normal capability negoitiations. has afi/safi pairs. uses SNPA portion of the MP_Reach_NLRI
[01:25:39] <ggm> can do tracing etc. on the virtual wires
[01:25:50] <ggm> AFI/SAFI. what home? IANA? IDR?
[01:26:23] <ggm> SAFI value set bit pattern. 0000 00LW <L> is presence of VPLS, <W> is the presence of VPWS.
[01:26:40] <ggm> SAFI value is set 3 to signify the NRLI encoding can be used for both
[01:26:47] <ggm> not overloading.
[01:27:11] <ggm> NLRI encoding kept same.
[01:27:29] <ggm> length/val for variable length stuff.
[01:27:46] <ggm> discovery example slide. see the doc.
[01:29:08] <ggm> virtual wires does the same stuff done in multicast. cannot go across exchange with same piece of pseudo-wire. not exactlu same, but similar
[01:29:29] --- javier has joined
[01:29:40] <ggm> provides tracing of the pseudo wires. you know even in packet where originated. doesnt expose policy from origin AS
[01:30:04] <ggm> solicit feedback. SNPA not generally used in many impl. what happens. real problem, solution with SNPA, and usage.
[01:30:24] <ggm> comment from <3?> (sorry, don't know the players)
[01:30:51] <ggm> draft has defined MLRI. has label. does auto disc using the NLRI. what defining here, same NLRI, no label inf in it.
[01:31:06] <ggm> Susan: L2 draft. refining existing drafts. ok if paul answers?
[01:31:43] <ggm> Paul. intention. yes, already draft, for discovery/signalling. this is for using LDP, to provide LDP with info to build pseudo wires.
[01:31:48] <Bill> <3?> is Rahul Aggarwal
[01:32:02] <ggm> Rahul. if already going to extent of doing this, makes sense to put label in.
[01:32:09] <ggm> Susan: feedback on mechanisms, not which is to be used
[01:32:17] <ggm> (ta for that Bill)
[01:32:31] <ggm> Rahul 2nd quick point, can use other spec, and NOT send the label info.
[01:32:41] <ggm> Susan. is this point on mechanism or usage?
[01:32:52] <ggm> Rahul missing something. mechanism & usage seem linked.
[01:32:57] <ggm> Kiri?
[01:33:07] <Bill> Kireeti
[01:33:31] <ggm> comment simple. nothing against doc. should do it. when get AFI, getting SAFIs, already one. want to use this AFI with SAFI 65 and
[01:33:38] <ggm> I'll tell you format. what you do here is perfect.
[01:34:04] <ggm> Kireeti. don't want to use mechanisms. I just want AFI. I'm happy
[01:34:15] <ggm> Susan see problem of tunnelling.
[01:34:30] <ggm> Kireete don't understand and dont care. does that work?
[01:34:36] <ggm> Susan. I'm not the SAFI administrator.
[01:34:54] <ggm> Gargi. same of different problems being solved by SAFI?
[01:35:16] <ggm> Kireeti sort of similar, solving auto discovery problem. but one does autodisc AND label.
[01:35:22] <ggm> Susan which we don't want to combine.
[01:35:36] <ggm> doesnt run into edge problem, don't understand but no.
[01:35:48] <ggm> Kireete. point you at doc.
[01:35:51] <ggm> Susan. ok
[01:36:07] <ggm> comments solicited.
[01:36:38] <ggm> Alex. early allocation
[01:36:56] <ggm> STD action in RTG area requires impls.
[01:36:59] <ggm> code point ranges use STD action policy.
[01:37:16] <ggm> implementers can't get codes from IANA. pick next == misuse of registry == BAD
[01:37:37] <ggm> oops. maybe I was channeling slides too soon.
[01:37:49] <ggm> alex message from yacob, let me restate more generally
[01:38:35] <ggm> risks collisions when IANA does the alloc.
[01:38:47] <ggm> seen it with RSVP flowspec, SENDER_TSPEC
[01:38:57] <ggm> IDR, BGP spec uses STD
[01:39:02] <ggm> proposal simple.
[01:39:25] <ggm> break tie from std and IANA action, allow early alloc in certain conditions. stable spec. adaquately documented, community interest
[01:39:29] <ggm> no change in semantics of field
[01:40:25] <ggm> WG chairs to gauge conditions. AD approves, chairs -> IANA. alloc temporary. 1 year, with renewal.
[01:40:33] <ggm> chairs responsible for deprecating expired codes and returning them.
[01:41:35] <ggm> if done this way, different to every group. tighten language to exclude MIBS
[01:42:47] <ggm> Kireeti. <missed>
[01:42:51] <ggm> susan agree bad for mibs
[01:43:13] <ggm> <mib> person want OID, want OID, have to keep saying 'cant have' doc which gives it, makes problem worse.
[01:43:17] <ggm> Kireeti. why say no OID?
[01:43:29] <ggm> susan if you've written a MIB, you'd know, changing OID is a pain
[01:44:11] <ggm> crowd at mike. no names.
[01:44:32] <ggm> nice to have had an OID we could have implemented in the draft, and not having to reimpl MIB
[01:44:58] <ggm> MIBlady. historically, if draft goes away, loose number in registry. if we want a decision like this, have to take and ops/mgt area
[01:45:20] <ggm> Zinin draft says that for existing registries requires IESG approval. default would not be able to apply. clear?
[01:45:33] <ggm> WG chair has to ask. IESG has to approve [lew]
[01:45:49] <ggm> in theory. MIBS aren't going to get them.
[01:46:34] <ggm> Zinin. mechanism, can be applied or not applied, depending if useful. understand OID in existing registry, without IESG approval not applicable. want to do, have to subit doc, through consensus process, and you woild say know. right?
[01:47:06] <ggm> Bill Fenner. strict policy about backwards compat. once OID allocation in MIB2 space, semantics don't change, ever. in this case, semantics change after year/2.
[01:47:32] <ggm> Zinin. if the codepoint remains allocated, cannot be changed. if deprecated, two years, not allowed to reuse. never.
[01:48:07] <ggm> Lew Berger allocations for temp. being 1 year, renewal way too restrictive. had things in ed Q for over 1 year. IESG approval for over 1 year.
[01:50:08] <ggm> Lew how to know when in IESG process
[01:50:11] <ggm> Zinin use tracker.
[01:50:44] <ggm> Dave Meyer. IANA can time anything out?
[01:50:51] <ggm> zinin no. make WG chairs responsible.
[01:50:57] <ggm> Dave have to replicate some piece of registry
[01:51:03] <ggm> Zinin repeating ML disussion.
[01:51:41] <ggm> Zinin periodic checks required.
[01:51:51] <ggm> Yacob. doesnt have to be 1+1. could be some other thing.
[01:52:35] <ggm> Gargi. what happens when lease not renewed, but code in use. permanently used
[01:53:01] <ggm> Zinin. need make make case, with IESG approval can prolong.
[01:53:04] <ggm> Susan. out of time
[01:54:28] <ggm> Kireeti idea currently deprecated for period of time, all impl get heck out. 2-3 years. want process to formalize. early allocs subject to constraints. if through, well and good. if not, give time period. do not want perm alloc by mistakes in small registries
[01:54:43] <ggm> <?> some ISPs run 8 year old cisco code
[01:55:47] <ggm> Zinin: mail to IANA cc 'please return code points'
[01:57:56] <ggm> Susan draft mechanisms/comments. use AFI/SAFI 'temp space' - then into pools -can do stuff to flag when using temp value. not policy heaven
[01:59:56] <ggm> WG chairs discussing mechanisms.
[02:00:36] <ggm> Zinin. concern I have: code stays for long time. cannot assume once value returned never ever see it on network
[02:00:56] <ggm> Yacob. upper part is probably experimental. 64 different numbers.
[02:02:03] <ggm> Lew alloc experimental codepoints at beginning of draft. all impls use codepoint to do favourite protocol, once standard codepoint,
[02:02:31] <ggm> all impls put standard in, send with std, accept both. old codepoint never changes, always stays in code
[02:02:36] <ggm> sorry Dino. not Lew. my bad
[02:02:42] <ggm> Bill Fenner
[02:03:23] <ggm> RFC3692. describes one method of allocating experimental numbers in IANA ranges. alloc small for experiments, configure to say for this experiment, use this experimental number. already documented. throw it into the mix
[02:03:35] <ggm> Yacob: dinos point emportant. its all codepoints.
[02:04:28] <ggm> Susan. charter discussions for going forward.
[02:05:01] <ggm> need some MIB discussions. not during these sessions. work outside of WG time, have MIB time for v2.
[02:06:26] <ggm> New work.
[02:06:50] <ggm> Fault isolation, Max prefix draft. ECMP. BGP collisions avoidance. Deterministic route redist. etc.
[02:07:08] <ggm> opening up the draft sequence again. offers accepted. throw charter open for discussion. want to hear about proposals for WG items
[02:07:19] <ggm> Yacob. not end of discussion. carries on on ML
[02:08:17] <ggm> [silence]
[02:08:30] <ggm> Yacob. describe importance of problem being solved. will need to order.
[02:08:37] <ggm> Susan still have to take docs off the top
[02:08:56] --- iljitsch has joined
[02:09:10] <ggm> Rudiger Volker.
[02:09:27] <ggm> top three on list. ECMP is maybe a bit high up.
[02:10:05] <ggm> DONE.
[02:10:18] --- ggm has left
[02:11:27] --- javier has left
[02:14:32] --- iljitsch has left
[02:17:56] --- Bill has left
[02:20:34] --- avri has left: Disconnected
[13:56:53] --- ignas has left: Replaced by new connection
[14:21:09] --- ignas has joined
[15:21:18] --- LOGGING STARTED
[18:35:38] --- ignas has joined
[20:02:38] --- ignas has left: Disconnected