[17:16:17] --- ggm has joined
[17:18:26] --- rpayne has joined
[17:20:54] <ggm> dave says "we'll give people a bit more time, to filter in, esp. from the routing area thing across the hall"
[17:24:10] <ggm> dave plugs the xml2rfc http://xml.resource.org/
[17:24:35] <ggm> Danny McPhersons doc on med considerations, ready for last call once a diagram is done.
[17:24:57] <ggm> Dave Meyer on collection communitities is waiting for AD input.
[17:25:08] --- avri has joined
[17:25:14] <ggm> embedded-addr is waiting for AD input
[17:25:45] <ggm> rift going to talk about it.
[17:25:55] <ggm> started after Vienna, called routing protocol overload
[17:26:05] <ggm> presupposed the answer to the Q it was supposed to be analysing
[17:26:13] <ggm> there was a design team, hasn't been too functional
[17:26:30] <ggm> tried a couple of approaches to this doc. current draft needs work, hasnt been touched for a couple of months
[17:26:37] <ggm> take a look at the sections which are controversial
[17:26:55] --- sakai has joined
[17:27:05] <ggm> fit for different kinds of signalling applications, decided to rewrite as inventory , not take position, couch as informational.
[17:27:10] <ggm> any readers [no]
[17:27:17] <ggm> Eric. still needed?
[17:27:34] <ggm> Dave Good question. purpose was to kill things going forward anyway. If we had an AD around, could ask 'em
[17:27:46] <ggm> Eric let it quietly expire., not going to make any difference
[17:28:16] <ggm> Yakov. support what Eric said. would like to ask WG chairs to let this draft expire, by WG consensus.
[17:28:29] <ggm> Dave take the sense of the room
[17:28:41] <ggm> Yakov I didn't mean in room I meant ML
[17:29:00] <ggm> Dave people read it, have to have read it, how many would like to see it die
[17:29:03] <ggm> [small number]
[17:29:10] <ggm> Dave how many wnt to send it to informational.
[17:29:19] <ggm> take to list as per Yakov
[17:29:48] <ggm> expired/inactive. bounded longest match. if dropped from WG, will take to personal submission
[17:29:59] --- avri has left: Disconnected
[17:29:59] <ggm> Dave who's read? [too few] ok Take to list
[17:30:11] <ggm> BGP-redist draft is the same. [not enough] take to List
[17:30:26] <ggm> Huston, BGP status report
[17:31:04] <ggm> Routing status report.
[17:31:19] <ggm> Dave and I thinking of making this a regular report. appreciate kind of reports you'd find useful
[17:32:59] <ggm> going through the history of route analysis, graph shape observations, entry of route views
[17:33:29] <ggm> the post route-views state shows banding, with about 40,000 entry span. high ones are inside-AS unagreggated entries. middle span is most common, lowest two are prefix filtered
[17:33:40] <ggm> Dave know which peers they are?
[17:33:50] <ggm> G lowest is NTT verio, highest is inside Telstra
[17:34:03] <ggm> Geoff cont
[17:34:26] --- avri has joined
[17:34:42] <ggm> hard to analyse. longer graph from '88 view of table size, see extended history 89-94 shows exponential growth, due to growth of B. then linear growth in CIDR, then sharp 99-2001, the 2001 boom was
[17:34:55] <ggm> interesting, lot of 'correction' with a drop in size. since then, best described as linear.
[17:34:59] <ggm> looking at last 12 months..
[17:35:41] <ggm> sharp points, notable discontinuities, where they go down, aggregation, up, transient means leakage, late June, was a consistent inject of more specifics, (graph from hourly snaps) overall view is back into linear growth in table size
[17:35:47] <ggm> V4 address span.
[17:36:17] <ggm> graph of total address space span. from feb 2000 to few days ago.- graph shows 1/4 of the space of the global net,
[17:36:48] <ggm> strong banding effects. bunch of distinct 161777260 entries == /8 flapping. two observations
[17:36:56] <ggm> 1) /8 flap 2) not everyone sees the same ones at the same time.
[17:37:12] <ggm> some folk playing routing games. variation in entire /8 visibility
[17:38:00] <ggm> overall shape best seen as single span. linear growth 2000-2002, across 2002, slower growth in addr consumption rates, since then has increased, over last 6 months relatively steep. # adds in table is relatively consistent linear, hard to say if continuing to grow
[17:38:17] <ggm> other part of table interesting, map routing entries is the more specifics. (of existing adverts)
[17:39:00] <ggm> in graph don't worry if it has more information, just trying to understand in SPACE terms gives no additional info. 2001, 55%. now 51%. not sharp decline
[17:39:09] <ggm> Dave done paths and prefixes?
[17:39:18] <ggm> Geoff, yes, dont have graph but similar
[17:39:27] <ggm> diff is fe w% points
[17:39:52] <ggm> pattern of adv. more specifics, as seen at route-views, once covered, leaks, adding no additional information. paths equal with more specific prefix
[17:39:59] <ggm> Unique ASNs. 2000-present
[17:40:06] <ggm> scale is 6000-18000 right now
[17:40:13] <ggm> around 17500 unique AS in use
[17:40:26] <ggm> majority originate some, very few transit only. some orig. more than others (!)
[17:40:39] <ggm> high slope, to march 2001, post 2001 linear, less steep.
[17:41:01] <ggm> systematic filtering or some other attribute going on here, 10 peers consistently see smaller number of ASns.
[17:41:31] <ggm> one, or two, consistently see many many much smaller, 1000 smaller. NTT/Verio does strong prefix filtering, when do, may not loos address info, some AS as more specifics, do not see any more
[17:41:48] <ggm> leads to observations, strong filtering may lead to information loss, open issue does it mean different paths.
[17:42:09] <ggm> avg AS path length. number of AS growing, shape of connectivity remain constant, AS path length should increase
[17:42:45] <ggm> across all peers of route-views, AS-PATHs, remove stuffing of around 4 ASs. end-to-end is 4AS across. as number of AS increases, interconnect density increases, paths tends to become higher
[17:43:19] <ggm> inference is, in distance-vector proto, exploring paths on withdraws, more exploration. adding more connectivity at each length, find on avg beyond 4, counting past the known internet., diameter is constant
[17:43:29] <ggm> v4 Aggregation potential
[17:43:45] <ggm> shows graph of relationship to agg path.
[17:44:07] <ggm> shows what would happen with removal of more specifics with same AS-PATH, but retaining its 'integrity' (ie same info in routing table)
[17:44:13] <ggm> only carry 98,000 in RIB.
[17:44:40] <ggm> if more aggressive, prune more, non-prepended, equal, more specific, take it out == 2000 more. so not a lot of prepending more specifics.
[17:45:13] <ggm> finally, same origin ignore path, 85,000 entries. so around 8-9000 entries with traffic engineering/differential annouce going on. bulk share path as aggregate
[17:45:23] <ggm> reasonable amount of more specifics with no inf.
[17:47:20] <ggm> [GGM Q's the state, Geoff clarifies]
[17:48:00] <ggm> Part of problem is not absolute size, but level of computation. if updates are more specifics flapping, then can argue yes. but if update rate is evenly distributed, its not so clear. cannot say at this stage
[17:48:07] <ggm> (could be real connectivity repair)
[17:48:18] <ggm> Now V6
[17:48:24] <ggm> slightly linear.
[17:48:34] --- jis has joined
[17:48:40] <ggm> number of routing entries not huge. starts at 400, now 640. over 03-04.
[17:48:49] <ggm> massive amounts of instability, leakage, breaks in connectivity
[17:49:09] <ggm> sure if looked at events, would see what made another 70 entries appear in v6 routing table
[17:49:22] <ggm> so appears to be small, and growth driven by external events as much as natural growth,
[17:49:26] <ggm> address span is hard to figure.
[17:49:32] --- Anders has joined
[17:50:08] <ggm> two sets of addresses being advertized with different params. RIR working 2001::/16. alloc unit is a /32, compared to 3ffe::/16 6bone, alloc a /24. most of what you see is biassed by 3ffe://16
[17:50:22] <ggm> dynamicisim is the moving of /24, rather than underlying 2001 work. wait and watch.
[17:50:26] <ggm> how many route?
[17:50:56] <ggm> 17,800 AS to v4. same measure in V6 shows 450. much closer to curve for number of entries, more 1:1
[17:51:15] <ggm> what we see is that if the goal in distribution method was strong aggregation between originAS and prefix, goal being met. strong aggregation
[17:52:03] <ggm> looking at agg possibilities, same technique in V4. can't strip many. around 80 more specifics, /48s. some /64s some random noise inbetween. in general v6 table doesnt display same issue
[17:52:13] <ggm> next views, look at entrire space as entity.
[17:52:48] <ggm> graph of entire address space. figure out inside each /8 bucket, how many adds still in IANA reserve pool., how many in RIR, how many to IETF, how much routed, how much not routed.
[17:52:49] --- suz-isc has joined
[17:52:58] --- bernt99 has joined
[17:53:26] <ggm> know addressspace not handed out, dominated by class A space. no effort to route, not being returned.
[17:53:39] <ggm> 20/8 to 30/8 and 4x/8 lots of unrouted.
[17:53:53] --- sadlerjon has joined
[17:53:54] <ggm> from 52/8 onwards, RIR allocation, most recent RIR space in class A, what is allocated is routed.
[17:54:09] <ggm> current RIR policies around routabilty, routing and assignment.
[17:54:29] <ggm> Historic B space is around 60% routed. Cs, older allocations, not much being routed compared to more recent.
[17:54:37] <ggm> AS numbers, same picture but in 256 AS blocks.
[17:54:50] <ggm> 1/2 way through. allocating around 33,000. dark green shows where RIR about to allocate.
[17:55:12] <ggm> old AS never die, just 'go away' -most recent RIR alloc AS have high routing potential, go further back and see fewer. AS dissapear.
[17:55:31] <ggm> V6, two /16 shown, the active ones. on the right is 3ffe/6bone.
[17:56:00] <ggm> bar is a /23, the current IANA->RIR alloc rate. on right its a /24 halfway, or all. this is why 6bone shows routing HUGE amount of space
[17:56:06] <ggm> appear to have checkerboarded their space
[17:56:51] <ggm> the Red in the RIR allocs, only small amount routed. bottom LHcorner. single large block allocated in May. the rest is alloc windows. not the same as V4, previous had complete space, this is sparce
[17:58:25] --- sakai has left: Replaced by new connection
[17:59:24] <ggm> Geoff shows movie of allocation history.l
[17:59:38] <ggm> can see rush on the B's in pre-CIDR times.
[17:59:56] <ggm> then introduction of RIR process, move to few /8 but mainly CIDR in old C space.
[18:00:01] <ggm> can see AS rate growing faster
[18:00:26] <ggm> internet boom didn't reflect address consumption./ RIR policy forces slower growth
[18:00:50] <ggm> can see class B routing very unstable, across entire address space
[18:00:55] <ggm> dancing /8s. flapping
[18:01:15] <ggm> 2001, see internet bust did slow down allocation rate, but AS numbers grew. density of internet was changing
[18:01:28] <ggm> see very rapid AS uptake, slower IP address
[18:02:10] <ggm> interested in feedback, useful pointer to information
[18:02:25] <ggm> ends.
[18:02:30] <ggm> Dave
[18:02:49] <ggm> planning on doing a routing summary at every meeting, whats happening with BGP.
[18:02:52] <ggm> thats the end of the Agenda
[18:03:46] --- bernt99 has left
[18:04:16] --- ggm has left
[18:10:16] --- sadlerjon has left
[18:10:58] --- suz-isc has left
[18:15:58] --- Anders has left
[18:18:30] --- avri has left
[19:14:16] --- jis has left
[19:35:02] --- rpayne has left