[01:52:09] --- torus has become available
[01:54:25] --- nm has become available
[02:00:44] --- brabson has become available
[02:01:43] --- nm has left: Replaced by new connection
[02:01:45] --- nm has become available
[02:01:47] --- nm has left
[02:02:03] --- Xag has become available
[02:02:05] --- nm has become available
[02:03:38] --- arifumi@jabber.org has become available
[02:03:46] --- jeffa has become available
[02:04:04] --- ggm has become available
[02:04:09] --- FDupont has become available
[02:04:09] --- brabson has left: Disconnected
[02:04:16] <ggm> charter is online. its baked. not being discussed today.
[02:04:25] <ggm> wanted 2hr slot., got 2x 1hr. doing 4 things today
[02:04:44] <ggm> kurtis will review the state of the nation first, then Arch, protocol, Crypto locaters, and Triggers.
[02:04:53] <ggm> Tomorrow, same time, same bat-channel,
[02:05:25] <ggm> doing the applicability work, a directions for WG discussion, overlap with MIPv6 and some l3Shim state management
[02:05:27] --- iljitsch has become available
[02:05:48] <ggm> not directly on-charter but intetresting
[02:06:04] <ggm> AOB call?
[02:06:08] <ggm> Kurtis tomorrow is room 352
[02:06:18] --- WH has become available
[02:06:25] <ggm> <Geoff is in the chair with Kurtis, Geoff doingt the schpiel thus far>
[02:06:36] <ggm> Kurtis. repeating slideset from GROW
[02:06:38] <ggm> short recap
[02:06:56] <ggm> checking people understand the problem [forest of hands]
[02:07:04] --- brabson has become available
[02:08:29] <ggm> Multi6 was NOT taskd to do proto devel. re-charter or form new WG, decided to form new WG. multi6 isn't closed down yet, still have items in RFC-ED Q. Shim6 is in INT area. doing devel work on the drafts coming out of Multi6
[02:08:38] <ggm> these have been re-submitted into Shim6.
[02:08:51] <ggm> minor changes. some new documents, applicabilty statements, will map into milestones.
[02:09:03] <ggm> milestones are aggressive. aiming to complete in the year [laughter]
[02:09:09] <ggm> first done: get drafts out.
[02:09:18] <ggm> aiming to complete applicability statement and arch by next IETF
[02:09:49] <ggm> so far, on track.
[02:10:35] <ggm> next steps. state mgt, ID characteristics, locater behaviours, state transitions
[02:10:41] <ggm> Thats it for the summary.
[02:10:46] <ggm> <done>
[02:12:18] <ggm> shim6-arch -draft-ietf-multi6-architecture-04.txt now in RFC pub Q. what multihoming was about, kinds of approaches considered in multi6
[02:12:19] --- mark.stripes@jabber.org has become available
[02:12:56] <ggm> Geoff talks about the approaches in this multi6 draft. covers SWOT, identities, roles.
[02:13:05] <ggm> mentions HIP 'ephemeral' identity
[02:13:12] --- mo7sen has become available
[02:14:45] <ggm> General Qs. How is a session/equiv state established?
[02:15:11] <ggm> what triggers this capability being negotiated? what packets? timers? per session? wait for locater problem? this set of Q is how to set up the equivalent state.
[02:15:42] <ggm> Shim6 view is 'restoration and repair' as the major view. this means 'what is the trigger' to make this repair happen. is the trigger up the stack, or is it down in the internet layer?
[02:15:53] <ggm> is it best-effort datagrams, or what?
[02:16:16] <ggm> next Q: what is an equivalence state. differentiate locator pools by session or class, its a subtle distinction compared to hosts. for every session between hosts.
[02:16:40] <ggm> how to select locaters. once a set is bad ,how long is it bad? month? year? 2secs? what refreshes?
[02:16:52] <ggm> at the IP level, when do you know its all ceased? what times out the state? what is a stale state.
[02:16:56] <ggm> generic Q's posed in the draft.
[02:17:04] <ggm> Aaron Falk.
[02:17:19] <ggm> saying Shim6 is restoration and repair, but I thought it was multihoming, load balancing, broader than that,
[02:17:45] <ggm> Geoff as I understand this, this is unicast, rather than multicast. not simul packets on all locaters, or even a subset. not load balancing, its path flicking
[02:17:50] <ggm> Aaron. alternate paths idle?
[02:18:03] <ggm> Geoff, at sesion level idle? maybe not before
[02:18:10] <ggm> Erik N. don't switch paths unless something wrong
[02:18:27] <ggm> Marcel. probably choose different ISPs
[02:18:46] <mark.stripes@jabber.org> Marcelo Bagnulo
[02:18:51] <ggm> Geoff, things don't change unless there is anevent at a host level. may find host talking to host at same site ondifferent locators. load balance is statistical, not planned
[02:19:30] --- resnick has become available
[02:19:41] <ggm> Jason Chiller. multi6 as replacement as replacement for V4 multihoming has to do more than failover. interesting TE cases used today in v4. primary backup, loadshare, not neeesarily routing based. I think multi6 should solve
[02:20:09] <ggm> <missed name> working in this for long time. failover is tricky part. get that right, can see limitations, easier to work on the other stuff, not core. extra
[02:20:30] <ggm> Geoff. grep TE in the document. will find discussion in the doc.
[02:20:46] <ggm> Jason have seen TE in the doc, around failure/degraded link, no text about primary backup or shortest path.
[02:21:42] <ggm> Arch draft. very incomplete.
[02:21:52] --- suz has become available
[02:21:52] --- kurtis has become available
[02:21:54] <ggm> endpoint id consideraitons, functional decompositoon.
[02:22:05] <ggm> thinking about adding equivalence state definition, design trade-offs.
[02:22:22] <ggm> eg differentiate between bulk TCP and a streaming UDP, or deal the same way?
[02:22:33] <ggm> enumerate the design decisions
[02:22:44] <ggm> the basic Shim6 ID/LOC split diagram.
[02:23:13] --- FDupont has left
[02:23:20] --- jeroen has become available
[02:23:29] <ggm> upper level identitiy. the first DNS resolution address, or however you get it. that is the 'sticky' label used by upper levels. mapped on the fly, packets coming down into different locaters. de-mapped at receiver end. become same upper level identity
[02:23:43] --- FDupont has become available
[02:23:45] <ggm> where in the stack? below IP endpoint sublayer. above IP routing.
[02:24:22] <ggm> defines the ULID. upper level identifier. 'its just an address' or if have multiples 'selection from set' -probably viable locater. drawn from name to address space, reverse mappable (if using in-addr.arpa)
[02:24:43] --- tskj has become available
[02:25:10] <ggm> better selection is deterministic. each one gets same. not neeccessarily the case. can be a reference context to pass to other people. inbetween hosts. its not ephemeral. its semi persistent. last as long as addresses do. not eternal, but not ephemeral.
[02:25:17] <ggm> Functional Decomposition.
[02:25:41] <ggm> don't assume the other end can do it. assume other end can't then do facility negoitiate, then exchg locater sets, do mapping if agree can do it mutually.
[02:25:44] <ggm> Maintaining State
[02:25:59] <iljitsch> ggm: "missed name" after Jason Chiller: Iljitsch van Beijnum
[02:26:09] <ggm> more work needed here. -how to detect its time to use something else at the IP level? no RTT/State that a session has.
[02:26:32] <kurtis> I think that is Jason Schiller BTW
[02:26:49] <ggm> do pseudo transport with heartbeat? pushed up into session or separate? what does transp. proto tell you about timeouts? if TCP is going to die, does it signal lower level, ask for locater shift to avoid dying.
[02:27:04] <ggm> {at times like these, spelling drops down the priority queue :-}
[02:27:27] <kurtis> ggm: we fix that in the minutes so no worries
[02:27:33] <ggm> Some apps, VOIP have no tolerence of multi-second delay. others don't care. so failover times matter app by app
[02:27:48] <ggm> may need different recovery mechanisms.
[02:28:14] <ggm> Possible Failure triggers. whats valid? the N**2 problem: if you have 10 locaters each end, its a big number of combinations to test.
[02:28:38] --- geir_egeland has become available
[02:29:27] <ggm> have to do state change. tell TCP rtt estimates wrong. reset congestion window. you are doing the change. the other end gets unexpected event. has to know from received packets trigger signals to upper levels , request RTT/congestion changes. signal things changing. active/passive ends.
[02:29:27] <ggm> Removing state
[02:29:27] <ggm> big Q whats the timeout point. risk of doing it too early. vertical signalling requires work
[02:29:27] <ggm> Open Issues
[02:29:54] <ggm> Use of hash based addrs, crypto Addrs in low 64 address needs to be considered. do you know in advance what locater set, or NOT know, will some be dynamic. has implications with HBAs etc.
[02:30:08] <ggm> exchanging sets. interaction with site exit routers. need to be defined in this and other documents.
[02:30:19] <ggm> ULID selection. how deterministic.
[02:30:21] <ggm> Common Issues
[02:30:47] <ggm> whats the vertical signalling to IP layer and session. may not have explicit sigs for estab. or teardown. may be many conversations between endpoints.
[02:31:03] <ggm> are there 'distinguished' locaters you always need. may need for bootstrapping. or constant probe to find?
[02:31:25] --- mark.stripes@jabber.org has left: Logged out
[02:31:38] <ggm> Persistence. is there a home locater and switch or start from any. can bdinings be delayed? wait for failure, or do capabilities 'jiust in case' how much JIT how much J-i-Case?
[02:31:44] <ggm> Aaron Falk.
[02:31:50] --- jschoenwae@jabber.org has become available
[02:32:13] <ggm> this isn't strictly related. ignorant of neighbour discovery. is it a solved problem that a host can acquire adds from two different v6 nets?
[02:32:47] <ggm> Christian. H. happens all the time. Its not unusual
[02:33:07] <iljitsch> i have 5 ipv6 addresses right now.
[02:33:23] <iljitsch> obtained automatically from the wireless network.
[02:33:41] <iljitsch> (2 are site-local and 2 6to4, though)
[02:33:56] <ggm> spencer. continuing detection of failures.
[02:34:09] <ggm> 'we are about to decide it has happened' is that what you are on?
[02:34:12] <arifumi@jabber.org> why do we have site-local here ?
[02:34:22] <ggm> Geoff. discover/establish addds time and wait
[02:34:30] <iljitsch> present from the host?
[02:34:30] <ggm> {why do we have 2002 mapped here?}
[02:34:45] <ggm> I don't have solution, its a dimension of the problem
[02:34:58] <ggm> Spencer question terminiology. failure somewhere, about to decide its true, that causes transition
[02:35:02] <ggm> geoff finished
[02:35:20] <ggm> <I have to go offline for a bit. maybe 10 min. sorry folks>
[02:36:06] <iljitsch> geoff: questions
[02:36:12] <iljitsch> errik nordmark is up
[02:36:39] --- dthaler has become available
[02:36:41] --- dthaler has left: Disconnected
[02:36:57] --- dthaler has become available
[02:36:57] <iljitsch> kurtis: feedback at shim6@ietf.xmpp.org
[02:37:03] --- rbless has become available
[02:37:05] <iljitsch> (jabber, nog email)
[02:37:10] --- dumdidum has become available
[02:37:13] <iljitsch> not
[02:37:44] <iljitsch> erik:
[02:37:55] <iljitsch> couple of drafts, details aren't worked out yet
[02:37:59] --- gih has become available
[02:38:17] <iljitsch> - no separate id space, ulids geoff talked about
[02:38:27] <iljitsch> (too fast for details)
[02:38:46] <iljitsch> dns assumptions: fqdn might be for a service or a host
[02:38:58] --- jijohna has become available
[02:39:06] <iljitsch> fqdn returns set of potential ulids, try them until one works
[02:39:22] <iljitsch> then pass locators back and forth with peer
[02:39:47] <iljitsch> desire to optimize failure during initial contact
[02:40:06] <iljitsch> deferred context establishment
[02:40:16] <iljitsch> don't penalize every connection with shim overhead
[02:40:21] --- yushun has become available
[02:40:34] <iljitsch> three eents: inital contact, decide to do shim setup, rehoming after failure
[02:41:10] <iljitsch> also need to handle failures during initial contact. let app handle this, optimization possible with shim?
[02:41:54] <iljitsch> pX = packet X in sequence
[02:42:03] <iljitsch> no state change at p1 for DoS reasons
[02:42:19] <iljitsch> if icmp error or no response to p1: no shim support
[02:42:26] <iljitsch> very similar to HIP exchange
[02:42:54] <iljitsch> no shim because peer apparently doesn't implmenet it when no answer for p1
[02:42:58] <iljitsch> changes in documents:
[02:43:19] <iljitsch> mostly editorila since multi6-l3shim-00
[02:43:25] <iljitsch> terminology changes
[02:44:03] <iljitsch> no changes to multi6-functional... just name change
[02:44:07] <iljitsch> open issues:
[02:44:25] <iljitsch> receive side deumlltiplexing (see list since ietf62)
[02:44:35] <iljitsch> state management: how/when remove state?
[02:44:49] <iljitsch> packet formats for control procol
[02:44:55] * FDupont has set the topic to: SHIM6 BOF
[02:45:04] <iljitsch> apis for ulp advice
[02:45:13] <iljitsch> path maintenance and exploration protocol
[02:45:17] <kurtis> /topic shim6 WG
[02:45:22] <kurtis> sigh
[02:45:34] <kurtis> /topic shim6 WG
[02:45:46] <kurtis> ok, I will pay attention to the slides again :-)
[02:45:54] <iljitsch> question for all of us: pick one approach, hash it out, see if we're on the right path
[02:46:11] <iljitsch> next steps?
[02:46:25] <iljitsch> use flow label to carry context tag
[02:46:37] <iljitsch> different flow label after locator change (number picked by receiver)
[02:46:57] <iljitsch> unilateral removal of shim6 state plus error message when no state to trigger peer re-establishment
[02:46:58] * FDupont has set the topic to: shim6 WG
[02:47:08] <iljitsch> pick control protoocol number
[02:47:18] --- dmm has become available
[02:47:30] <iljitsch> alternative: explore 8 byte extension header for data packets after failover
[02:48:26] <iljitsch> sorry missed pekka's comment
[02:48:26] --- jeroen has left: Replaced by new connection
[02:48:39] --- jschoenwae@jabber.org has left
[02:49:30] <iljitsch> anononymous jim bound: as long as we don't swap ulid, we've broken the tcp api, lot of this is in sctp, question: are we at a point in arch discussion...
[02:49:35] <iljitsch> are we breaking tcp api?
[02:49:56] <iljitsch> what would need to change in sctp ? telcos (?) are asking for sctp
[02:50:12] <iljitsch> a lot of this exist, why use the energy, unless we can dynamically change ulid.
[02:50:41] <iljitsch> erik: don't change ulid, only addresses on the wire, rewritten by receiver so tcp doesn't see any changes
[02:50:45] --- mark.stripes@jabber.org has become available
[02:50:52] --- pierz has become available
[02:51:10] <iljitsch> jim: swapping out ulid is pretty intense
[02:51:50] <iljitsch> kurtis: we don't know, had this discussion about sctp in shim6 and other overlaps, but shim6 must also supply functionality for tcp and udp
[02:52:10] <iljitsch> jim: these are the family jewels, we're talking about changing a lot of stuff
[02:52:16] <iljitsch> pekka nikkander:
[02:52:32] <iljitsch> re talk in internet area meeting
[02:52:55] <iljitsch> (missed some)
[02:53:13] <iljitsch> we need to figure out the relation shim6 sctp
[02:53:33] <iljitsch> i don't understand why need anything in the ip header at all
[02:54:28] <kurtis> FYI: We are doing the other two documents tomorrow so we will let this discussion go on...
[02:54:39] <kurtis> and I will say that at the mike as soon as I get a chance :-)
[02:54:45] <iljitsch> erik: receiver needs to undo sender changes to the paccket correctly. what information does the receiver need for this?
[02:55:06] <iljitsch> interest in having sessions with different ulids between the same hosts.
[02:55:17] <iljitsch> christian:
[02:56:01] <iljitsch> jim and pekka pointed out that there is a lot of interaction between this and transport protocols, mistake if we don't explore potential here
[02:56:18] <iljitsch> need to look at what could and should change in transports
[02:56:25] <iljitsch> i have to stop now
[02:56:50] <ggm> <I can take it back>
[02:56:52] --- hardaker has become available
[02:57:06] <ggm> take SCTP take requirements, see if SCTP can meet them, see whats missing. I can talk about this.
[02:57:28] <ggm> Erik. not saying do instead. have SCTP out there, how do they fit together. and with TCP.
[02:57:45] <ggm> <x> SCTP has own API, have to change appls. Christian was talking about changing transports.
[02:58:02] <ggm> But billions of apps. feasability issue
[02:58:17] <ggm> Erik. discussion happened in the past. we have apps using SCTP api for mh. what happens when shim comes in.
[02:58:27] --- hardaker has left: Disconnected.
[02:58:41] <ggm> Michael -run from SCTP perspective, its single homed, can use shim to do stuff. don't have control of this within SCTP.
[02:58:55] <ggm> Erik may be sub-optimal
[02:59:15] <ggm> pekka nickander.
[02:59:21] <iljitsch> that was me at the mike after i said i had to stop
[02:59:54] <ggm> don't know how deep to go here, but as impl option, may annotate, upper lay ID belongs to, pass as-is with locaters to SCTP. as SCTP api processing, replace locater with ul ID
[02:59:59] <ggm> different ops depending
[03:00:07] <ggm> {thanks Iljitch}
[03:00:12] <ggm> {sorry I missed your name!}
[03:00:34] <ggm> case when start to impl, different i/f between shim6, tcp, udp depending on what want to achive.
[03:00:48] --- geir_egeland has left
[03:01:07] <ggm> if this works out, gets to arch with separtated loc/id, need to reconsdier location of congestion control. separate functions for single path vs congestion e2e or connections. research
[03:01:10] <ggm> michael. thats research
[03:01:23] --- brabson has left
[03:01:55] <ggm> Christian sending packets to diff addresses in SCTP is easy. making sure not spoof on receive is harder.
[03:02:19] <ggm> in case of UDP, its not a protocol its a pass-through, the proto is in the application,. its not APIs that matter, info has to go right up to app.
[03:02:42] --- mark.stripes@jabber.org has left
[03:02:45] <ggm> Tomorrow, will cram into 1 hour 70 mins. will do crypto locaters and triggers tomorrow, have time for discussion then.
[03:02:54] <ggm> I hope discussion has teased out the issues.
[03:02:59] <ggm> Kurtis start 9 sharp.
[03:03:01] <ggm> Geoff: close
[03:03:04] --- FDupont has left
[03:03:05] <ggm> <done>
[03:03:06] --- jijohna has left
[03:03:08] --- Xag has left
[03:03:08] --- jeffa has left
[03:03:18] --- iljitsch has left: Disconnected
[03:03:19] --- torus has left
[03:03:38] --- nm has left
[03:06:53] --- dthaler has left
[03:11:12] --- arifumi@jabber.org has left: Logged out
[03:13:31] --- tskj has left
[03:13:39] --- kurtis has left
[03:15:46] --- suz has left
[03:17:59] --- gih has left
[03:20:51] --- resnick has left: Disconnected
[03:21:19] --- WH has left: Disconnected
[03:21:38] --- dumdidum has left: Disconnected
[03:22:04] --- mo7sen has left: Disconnected
[03:22:09] --- rbless has left
[03:22:24] --- pierz has left: Disconnected
[03:28:34] --- ggm has left: Disconnected
[03:37:08] --- dmm has left
[03:37:15] --- yushun has left
[03:37:52] --- WH has become available
[03:37:53] --- rpayne has become available
[03:57:31] --- jeroen has become available
[04:03:26] --- rpayne has left
[04:54:28] --- jeroen has left
[05:19:36] --- WH has left: Disconnected
[07:03:14] --- gih has become available
[07:07:58] --- WH has become available
[07:20:02] --- gih has left
[07:39:58] --- WH has left: Disconnected
[07:57:32] --- gih has become available
[08:16:59] --- gih has left
[08:38:58] --- gih has become available
[08:49:08] --- gih has left
[09:36:30] --- gih has become available
[10:19:31] --- WH has become available
[10:19:48] --- WH has left
[10:26:08] --- gih has left
[10:58:15] --- gih has become available
[11:33:40] --- gih has left
[16:04:27] --- gih has become available
[18:24:25] --- gih has left
[23:51:19] --- gih has become available