2005-05-05 [08:52:34] * jhma has joined the channel. 2005-05-05 [09:40:39] * jhma quit IRC altogether 2005-05-05 [12:38:31] --- dr has become available 2005-05-05 [13:21:06] --- iljitsch has become available 2005-05-05 [13:44:40] --- rhe has become available 2005-05-05 [13:54:49] * jluk has joined the channel. 2005-05-05 [13:55:36] * marcoh has joined the channel. 2005-05-05 [13:56:10] * UncleH has joined the channel. 2005-05-05 [13:57:21] * geertn has joined the channel. 2005-05-05 [14:04:39] --- slz has become available 2005-05-05 [14:04:40] --- jhma has become available 2005-05-05 [14:07:07] * KnockandO has joined the channel. 2005-05-05 [14:07:31] * uk has joined the channel. 2005-05-05 [14:07:56] --- ferenc has become available 2005-05-05 [14:08:11] --- iljitsch has left 2005-05-05 [14:08:11] --- iljitsch has become available 2005-05-05 [14:08:11] --- iljitsch has left 2005-05-05 [14:08:11] --- iljitsch has become available 2005-05-05 [14:09:53] --- fil has become available 2005-05-05 [14:10:31] * prt has joined the channel. 2005-05-05 [14:11:01] --- EamonnMcGuinness has become available 2005-05-05 [14:11:06] --- EamonnMcGuinness has left: Logged out 2005-05-05 [14:11:40] --- EamonnMcGuinness has become available 2005-05-05 [14:13:25] I'll be taking questions for this WG 2005-05-05 [14:15:09] * JC has joined the channel. 2005-05-05 [14:15:16] --- geoff has become available 2005-05-05 [14:15:24] Portable has nothing to do with anycast. 2005-05-05 [14:15:42] --- geoff has left 2005-05-05 [14:16:25] --- fil has left: Replaced by new connection 2005-05-05 [14:16:30] --- fil has become available 2005-05-05 [14:19:18] it IS PI... just that you need to pay RIPE for it ;) 2005-05-05 [14:19:38] (directly, in form of membership) 2005-05-05 [14:19:38] dr: what are you talking about? 2005-05-05 [14:19:43] It is PI, but called PA :) 2005-05-05 [14:20:22] slz: correct 2005-05-05 [14:20:34] Is it really PI if you are the "P"? :) 2005-05-05 [14:20:45] rhe: you don't need to be a P 2005-05-05 [14:20:51] (marcoh) is there actually a difference if everybody gets a /32 ? 2005-05-05 [14:21:07] rhe: look at all the enterprise LIRs 2005-05-05 [14:21:19] what about them? 2005-05-05 [14:21:33] (marcoh) wether you call it PI or PA, doesn't metter for the min-alloc size and as such won't have a hit on the global table 2005-05-05 [14:21:45] Well i don't really care if i receive a /32 or a /40 or a /48 ... It's just a routing table entry 2005-05-05 [14:21:57] (marcoh) in fact, in the long run it might be more controllable 2005-05-05 [14:22:33] ah there we have it, "Stealth PI" :) 2005-05-05 [14:22:49] slz: yeah, but the 200-customer rule wasn't effective in stopping that anyway 2005-05-05 [14:22:52] slz: look at Google 2005-05-05 [14:22:56] Correct, i know it :-> 2005-05-05 [14:22:59] (marcoh) slz: giving people a /32 might imrpove things 2005-05-05 [14:23:13] improve what? 2005-05-05 [14:23:17] marcoh: depends in which way you think, it will improve things... 2005-05-05 [14:23:22] (marcoh) II would rather see alll /32's instead op people trying to multihome a /40 2005-05-05 [14:23:34] I don't understand that. 2005-05-05 [14:24:12] Well i do multihoming with /40s .. but i also can open up two more LIRs and announce two more /32s instead..where's the difference 2005-05-05 [14:24:19] (marcoh) well, in the end...if we stick to the 200 rule and prevent large corps from getting their own /32, somebody will start announcing more specifics 2005-05-05 [14:24:29] didn't you listen to geoff yesterday? 2005-05-05 [14:24:56] let's not have the Great Multihoming Debate here pls 2005-05-05 [14:25:00] :-) 2005-05-05 [14:25:10] won't lead anywhere 2005-05-05 [14:25:12] well it's connected to that in some way 2005-05-05 [14:25:26] sure it is, but we are not really supposed to discuss things here 2005-05-05 [14:25:27] but that's why I went through all the trouble with the unreachable windows media streams and jabber servers 2005-05-05 [14:25:38] (marcoh) it''s not very likely that somebody running a /40 from a certain LIR, will renumber his entire network when he switches provider 2005-05-05 [14:25:55] your point being? 2005-05-05 [14:26:14] (marcoh) give them their own alloc 2005-05-05 [14:26:26] I have a comment: we still don't see the difference between personal opinion and fact in this discussion. 2005-05-05 [14:26:27] (marcoh) then we now what to expect 2005-05-05 [14:26:58] In any case, the 200 assignments rule is very unproductive.. i doubt that anyone will fulfill the rule within the timeframe 2005-05-05 [14:26:59] sure give one to everyone, make it all portable, create a new layer below ip that does the actual routing. hey wait we already have that, it's called dns. 2005-05-05 [14:27:28] if you can't even imagine having 200 customers in 2 years why are you in business in the first place??? 2005-05-05 [14:27:44] i have >5000 customers, i have ~10 that want Ipv6 2005-05-05 [14:27:58] (marcoh) iljitsch: i'm certainly not giving each dsl cust a /48 2005-05-05 [14:28:07] why not? 2005-05-05 [14:28:11] (marcoh) they should be lucky with a /64 2005-05-05 [14:28:15] customers == business customers only here ATM ... 2005-05-05 [14:28:23] I'm glad I'm not your customer. 2005-05-05 [14:28:30] I have 3 /48s by the way 2005-05-05 [14:28:58] (marcoh) iljitsch: your not the average customer 2005-05-05 [14:29:14] Well i'd happyly assign /48 to every signle customers, if there were any more requsts at all :-) But there aren't ... 2005-05-05 [14:29:19] (jluk) iljitsch: Why three ?? 2005-05-05 [14:29:20] no but why do you want to squeeze addresses out of your customers? give them a /48 already. 2005-05-05 [14:29:23] (marcoh) I have > 40k people who hardly know where the powerbutton is located 2005-05-05 [14:29:45] one for my old tunnel, one from my new isp native for my server, one for my adsl 2005-05-05 [14:30:28] --- irc has left: Replaced by new connection 2005-05-05 [14:30:28] --- irc has become available 2005-05-05 [14:30:30] --- irc has left: Disconnected 2005-05-05 [14:30:31] --- irc has become available 2005-05-05 [14:30:38] (jluk) marcho - one of your customers found the powerbutton ?? 2005-05-05 [14:30:42] --- damien has become available 2005-05-05 [14:30:42] I'm not saying i need it, but I don't see a really good way to do it differently, splitting a /48 between server and adsl that aren't connected doesn't make sense. 2005-05-05 [14:31:02] you don't 2005-05-05 [14:31:47] you might have a /64 on your housing VLAN at some colo ISP and your /48 at home 2005-05-05 [14:32:02] with the /64 not being a more-specific of your home /48 2005-05-05 [14:32:20] * UncleH has joined the channel. 2005-05-05 [14:32:36] it's acutally pretty hard to come up with a usable provisioning model for this stuff. 2005-05-05 [14:32:39] at least that's my interpretation of things I recommend to people. 2005-05-05 [14:33:18] I recommended to this ISP (they're also my customer) to give every colocated customer a /48 automatically because it's simpler. 2005-05-05 [14:33:19] * StoCron has joined the channel. 2005-05-05 [14:33:28] This is a hen-and-egg problem.... if you can't get an allocation, you won't start assigning Prefixes to customers <...> 2005-05-05 [14:33:30] yes 2005-05-05 [14:33:39] and put the first /64 out of this /48 onto the VLAN 2005-05-05 [14:33:59] I helped to customers with their /32s, both got them within 48 hours no questions asked. 2005-05-05 [14:33:59] if they use firewalls and other routed structure in the colo VLAN, route the rest to that 2005-05-05 [14:34:00] (marcoh) iljitsch: as a general rule, we don't encourage people to do their own routing at colo 2005-05-05 [14:34:12] (marcoh) only the bigger customers have routers installed 2005-05-05 [14:34:35] * KnockandO has joined the channel. 2005-05-05 [14:34:36] * prt has joined the channel. 2005-05-05 [14:34:51] everything that I've ever seen running v6 is a router. :-) 2005-05-05 [14:34:58] (even windows xp) 2005-05-05 [14:35:46] *applause* 2005-05-05 [14:37:04] What i still don't get is why there is no "200 assignment rule" for Ipv4 allocations if people think they need it for Ipv6 allocations :-) 2005-05-05 [14:37:49] * [rvs] has joined the channel. 2005-05-05 [14:37:51] because ipv4 is as good as dead we need ipv6 to remain viable for the next 30 to 50 years so it's way too soon to screw it up now. 2005-05-05 [14:37:54] and this is not funny. 2005-05-05 [14:38:21] (marcoh) as a procedural argument, would be usefulll if people state their name and affiliation when walking to the mic 2005-05-05 [14:38:22] Basically you guys are saying "we want our class A even though there are only 127 of those" 2005-05-05 [14:38:37] this is thomas narten from ibm. 2005-05-05 [14:38:39] (StoCron) marcoh: ok 2005-05-05 [14:38:51] right on thomas! 2005-05-05 [14:39:00] listen to him! 2005-05-05 [14:39:07] (marcoh) Sto: I''m in the room,, but not everybody is 2005-05-05 [14:39:09] I'm sure we will see IBM having it's own allocation ;) 2005-05-05 [14:39:35] and what are they going to do with it? Use it in 1 of their 500 offices? 2005-05-05 [14:39:50] (marcoh) dr:: and what is wrong with that, they are far bigger as the ISSP's who hold a /32 2005-05-05 [14:39:59] no, announce more-specifics all around 2005-05-05 [14:40:02] --- Fearghas has become available 2005-05-05 [14:40:11] (marcoh) have them route it themselves and announce the aggregate 2005-05-05 [14:40:14] but they don't have 200 customers to assign /48s to? 2005-05-05 [14:40:14] so you're saying they should have 500 prefixes? 2005-05-05 [14:40:21] that would increase the v6 table by 40% 2005-05-05 [14:40:40] marcoh: it's always easy to argue not giving things to OTHER people when you will have no problems getting them :) 2005-05-05 [14:40:48] marcoh: well that would have worked before they sold their world wide network to AT&T 2005-05-05 [14:40:51] iljitsch: I just say that I guess that this MIGHT happen 2005-05-05 [14:40:52] (marcoh) dr: give the large once a /32 and everybody can filter as you don't expect any more specifics 2005-05-05 [14:41:08] they won't accept ISP-provided PA space for their offices 2005-05-05 [14:41:24] marcoh: there is no requirement to announce the aggregate 2005-05-05 [14:41:40] why don't you guys accept that the ietf couldn't make this work after 10 years of discussions so move on. 2005-05-05 [14:41:43] (marcoh) we might want to add one :) 2005-05-05 [14:41:48] a suggestion, yes. a stated intend, yes. a requirement? no. and for sure no enforcement 2005-05-05 [14:42:22] (marcoh) dr: why not loosen the allocation policy but require the aggregate being present 2005-05-05 [14:42:34] (marcoh) as a compromise 2005-05-05 [14:42:51] (marcoh) and we all filter at /32 boundaries 2005-05-05 [14:42:52] what do you want to aggregate? 2005-05-05 [14:43:23] (marcoh) ijitsch: force people to not break up allocs in more specifics 2005-05-05 [14:43:40] that just raises the price point 2005-05-05 [14:43:43] (marcoh) but be more liberal in to handing out alloc's to larger non-LIR's 2005-05-05 [14:43:51] (KnockandO) The word "force" sounds kind of hollow when you cannot exert pressure. 2005-05-05 [14:43:54] companies will open a LIR for every (larger) office then 2005-05-05 [14:44:03] if everyone with 1 customer can get a /32 then they don't need to deaggregate in the first place so this is meaningless. 2005-05-05 [14:44:07] (UncleH) as they do already iirc 2005-05-05 [14:44:10] it's obviously VERY easy to make up "I'm not an end site"... see Google 2005-05-05 [14:44:19] see the CH school 2005-05-05 [14:44:20] (KnockandO) And second, Marcoh, we're not always talking about non-LIRs here. There are LIRs in the room that cannot get an alloc. 2005-05-05 [14:44:28] see Akamai 2005-05-05 [14:44:32] yes, just peer worldwide and you're halfway there... 2005-05-05 [14:44:33] (UncleH) uunet/mci/whatever already have som 21 LIRs in the riep region 2005-05-05 [14:44:34] sure, everyone can say "i have customers!" 2005-05-05 [14:44:35] (StoCron) yep, like big transit providers 2005-05-05 [14:44:39] see [add several more] 2005-05-05 [14:44:47] (StoCron) dr: Akamai is a bad example, they regionalize by DNS anyway 2005-05-05 [14:44:48] (KnockandO) dr: The CH guys made it very plausible to me. 2005-05-05 [14:44:49] it's currently just a question of money and some lying 2005-05-05 [14:45:16] StoCron: so for what do they have the /32 then? 2005-05-05 [14:45:24] all I'm hearing is "the 200 rule is bad! throw it out!" but I don't hear any reasonable alternatives. Or any alternatives, really. 2005-05-05 [14:45:47] (StoCron) dr: who? Akamai? Dunno. *One* /32 is not really sufficient for Akamai's network model anyway 2005-05-05 [14:45:50] StoCron: they are a clear end site 2005-05-05 [14:45:51] (marcoh) iljitsch: alternative text is in the proposal 2005-05-05 [14:45:53] i still don't get why not every LIR should get an Allocation... sorry probably i'm dumb 2005-05-05 [14:45:55] (KnockandO) All I hear is Iljitsch trying to keep v6 to an elite group. 2005-05-05 [14:46:02] Stocron: they don't assign any IP space to "other organizations" 2005-05-05 [14:46:06] (KnockandO) That's not the way to let it really move forward. 2005-05-05 [14:46:12] (StoCron) slz: yep 2005-05-05 [14:46:17] slz: that's "PI for enough money" 2005-05-05 [14:46:32] dr: ..it's a start (as you said, no multihoming debate here :-) 2005-05-05 [14:46:39] stocron: so what do you propose we do for people who need/want more than one /32 (for other reasons than having more than 50k /48 customers) 2005-05-05 [14:46:44] slz: so don't start it ;) 2005-05-05 [14:46:47] (KnockandO) Admittedly, Iljitsch, a lot of control is lost with softer policies. But after all, people would have the chance of using the stuff... 2005-05-05 [14:46:47] grrr 2005-05-05 [14:46:52] (StoCron) dr: you already have "PI for enough money and a small lie", that's a non-argument 2005-05-05 [14:47:06] * andy has joined the channel. 2005-05-05 [14:47:06] StoCron: yup 2005-05-05 [14:47:09] (StoCron) iljitsch: where's the problem with those with large networks? The policy is here and working 2005-05-05 [14:47:17] StoCron: I don't oppose dropping the 200-rule idiocity 2005-05-05 [14:47:22] (StoCron) ah, for *small* networks 2005-05-05 [14:47:27] (marcoh) on a personal view I would rather see 50k /32's in the global table opposed to the current mess we call the IPv4 DFZ (at 160k prefixes) 2005-05-05 [14:47:31] asking people wether they PLAN to have 200 customers is... ridiculous 2005-05-05 [14:47:31] KnockandO: yes, of course, IPv6 is SO much nicer without all the hoi polloi, I want to keep it to me and my friends. 2005-05-05 [14:47:48] (nonsense of course: you don't need a /32 to run v6) 2005-05-05 [14:47:49] marcoh: of course... but noone sees that point, or all ignore it 2005-05-05 [14:47:55] (KnockandO) Iljitsch: It burns down to kind of this, yes. 2005-05-05 [14:48:05] (KnockandO) And of course you don't need a /32. Nobody said that. 2005-05-05 [14:48:05] dr: what's so ridiculous about it? 2005-05-05 [14:48:17] (marcoh) slz: I think it's the ignoring 2005-05-05 [14:48:23] knock: what's your real name? 2005-05-05 [14:48:29] (KnockandO) But we had this again: You cannot interfere with somebody else's operations. 2005-05-05 [14:48:37] (KnockandO) Iljitsch: Elmar (from DENIC) 2005-05-05 [14:48:40] iljitsch: it's all in my book ;) 2005-05-05 [14:48:43] (KnockandO) (forgot you cannot see the whois) 2005-05-05 [14:48:53] (just kidding *G*) 2005-05-05 [14:48:56] dr: which book? 2005-05-05 [14:49:14] iljitsch: my book describing the construction of a crystal ball 2005-05-05 [14:49:26] (KnockandO) dr: Is a prerelease available? 2005-05-05 [14:49:37] Somehow i doubt that there will be a resolution and we discuss that again at RIPE100 2005-05-05 [14:49:40] yes I wrote a book that's something like that too 2005-05-05 [14:49:41] (KnockandO) (and btw, I like people talking about what might happen in 50 years...that was quite something there) 2005-05-05 [14:49:46] knockando: no, I'll keep the holy grail for myself :] 2005-05-05 [14:49:54] (KnockandO) dr: Buggrit. 2005-05-05 [14:50:03] (marcoh) guys: before it gets all too personal, this is archived 2005-05-05 [14:50:05] ko: yes, i really like that, too... REALLY .. *cough* 2005-05-05 [14:50:12] http://www.amazon.com/exec/obidos/ASIN/0596002548/ref%3Dnosim/muada-20/103-3913245-6650238 2005-05-05 [14:50:15] marcoh: nah, we're easy :-) 2005-05-05 [14:50:45] (KnockandO) marco: I hope there's nothing personal around here, it's just the usual opinion clash. Which is archived in the mailing list archives, anyway... 2005-05-05 [14:50:58] ACK 2005-05-05 [14:52:47] the discussion (in the public) could use some more open words at times. political correctness only gets you so far, but effectively hinders to discuss some underlying issues and agendas quite effectively. which is part of the problem why things move only slowly or even not at all. 2005-05-05 [14:53:40] we should really get some people with different viewpoints in a room and hash it out. 2005-05-05 [14:54:04] (KnockandO) Actually, I'd really have liked you, Iljitsch, to be here... 2005-05-05 [14:54:12] contrary to what you may think I really want this to work better than it does now, but just giving everyone a /32 isn't the solution. 2005-05-05 [14:54:35] (KnockandO) It breaks always down to people doing "sensible" things. 2005-05-05 [14:54:39] I was in Manchester. :-) 2005-05-05 [14:54:42] (KnockandO) And "sensible" is kind of a variable. 2005-05-05 [14:54:47] And 99% sure be at the next one. 2005-05-05 [14:54:50] (marcoh) let's go for the intermediate solution and try it ? 2005-05-05 [14:54:53] And at the next IETF. 2005-05-05 [14:55:06] (KnockandO) Iljitsch: I meant, in regard of current discussions. It's a very important time now 2005-05-05 [14:55:43] (KnockandO) Marco: I myself have this damn problem to solve, and I don't want to lie... 2005-05-05 [14:55:52] (marcoh) I mean, the global table can grow a bit 2005-05-05 [14:55:57] Any of you guys coming to http://www.whatthehack.org/ ? Then we can hash it out there. :-) 2005-05-05 [14:56:02] (KnockandO) So if we could get a fix on #alpha proposal, either yes or no, I'd know what to do next... 2005-05-05 [14:56:33] (KnockandO) The global v6 table is negligible compared to the v4 one, _BUT_ there are some vendor problems still, like no hardware support etc. 2005-05-05 [14:56:48] (KnockandO) Otherwise, simply blow it up to 10K prefixes, nobody would technically care. 2005-05-05 [14:57:24] (marcoh) KnockandO: I think most hardware support issues break performance, not the actual BGP process 2005-05-05 [14:57:28] 10% of all v4 prefixes are responsible for 65% of my traffic. 2005-05-05 [14:57:58] (KnockandO) marco: Yup, but CPU issues impact stability... 2005-05-05 [14:58:09] (marcoh) and if people would actually start doing some volume on v6 it might push vendors to fix things and managers to upgrade the network 2005-05-05 [14:58:16] (KnockandO) iljitsch: You're obviously using too many more specifics :) 2005-05-05 [14:58:22] note btw that cpu technology has run into some trouble the past few years. 2005-05-05 [14:58:44] (KnockandO) marco: It#s the chicken-and-egg problem again. No services - no customers - no vendors. But: no customers - no services - again, no vendors. 2005-05-05 [14:58:46] We really need MMX/SSE/Altivec BGP implementations. :-) 2005-05-05 [15:00:55] (StoCron) run OpenBGPd on your J-router 2005-05-05 [15:01:15] StoCron: stop baiting ;) 2005-05-05 [15:01:16] Pah, J's dont even have an altivec. 2005-05-05 [15:01:28] (KnockandO) Sto: Let's for a moment imagine people using real operating systems... 2005-05-05 [15:01:44] (StoCron) iljitsch: they don't need one, the thing is fast enough 2005-05-05 [15:01:45] (marcoh) KnockandO: like ? 2005-05-05 [15:01:51] I should compile Zebra with GCC4 on my mac, see if it autovectorizes. :-) 2005-05-05 [15:02:02] (KnockandO) marco: There's more in the world than a blown-up fish. 2005-05-05 [15:02:10] (StoCron) mentioning Zebra in a serious BGP discussion should be a crime... 2005-05-05 [15:02:27] (KnockandO) sto: you're right here :) 2005-05-05 [15:02:28] (marcoh) Sto; I second that 2005-05-05 [15:02:43] (KnockandO) And still, there are people running that as their core route servers. 2005-05-05 [15:02:51] (StoCron) the openbgpd really has some very good ideas behind 2005-05-05 [15:03:13] (StoCron) (and some short-sighted ones) 2005-05-05 [15:03:16] (KnockandO) Yup, but I asked about portability, and the answer was "not interested". So there it goes. 2005-05-05 [15:03:21] so what does it do that other stuff doesn't? (may send me this private if you want) 2005-05-05 [15:03:31] (KnockandO) The concept looks good, get someone with BGP experience to the team. 2005-05-05 [15:03:59] (KnockandO) (and I mean someone unlike me, someone with real clue...) 2005-05-05 [15:04:07] knock: :-) hm that couldn't hurt if they don't have someone like that now. 2005-05-05 [15:04:14] (StoCron) KnockandO: the problem is that there is no "unix" standard to talk to the kernel routing table, and that sucks big time - but is not a OpenBGPs problem, but mainly a "there are too many unixes out there" problem 2005-05-05 [15:04:20] (StoCron) not even the BSDs are compatible there 2005-05-05 [15:04:37] there are only four bsds, not too bad. 2005-05-05 [15:04:55] (KnockandO) Sto: There's a bunch of portable software, so one could at least show interest in how that's done. But then, OpenBSD/Open* has always been a bit too religious... 2005-05-05 [15:05:03] (StoCron) yah, but the kernel routing stuff is different between them all 2005-05-05 [15:05:17] even in the *bsd family? 2005-05-05 [15:05:23] (KnockandO) Yup. But I don't see why one couldn't separate routing and forwarding layers. 2005-05-05 [15:05:29] (StoCron) from what the openbgpd people say, yes 2005-05-05 [15:06:12] (KnockandO) sto: Then there's a lot of truth in it - they know the relevant systems quite well. 2005-05-05 [15:06:34] (KnockandO) (I suspect them of using Linux actually...) 2005-05-05 [15:06:58] yeah like that's a good idea... networking on linux is like fliying a pig 2005-05-05 [15:07:16] (KnockandO) With enough thrust... 2005-05-05 [15:07:21] right 2005-05-05 [15:07:30] but the navigation is a bit hard 2005-05-05 [15:07:33] (KnockandO) (btw, at least FreeBSE zeroes out, too) 2005-05-05 [15:07:42] bsd is SO much better that it's not even funny anymore 2005-05-05 [15:09:14] (KnockandO) Last time I checked, forwarding through the stack was faster on a Linux. That was 18 months ago. But then, I'm not happy with PCs forwarding, anyway. 2005-05-05 [15:09:57] that's the thrust part. the configuration just doesn't make sense and many less common features are implemented very strangely if at all. 2005-05-05 [15:10:35] (StoCron) the iproute2 stuff is weird, but seriously cool 2005-05-05 [15:11:05] why can't they use ifconfig like normal people? 2005-05-05 [15:11:29] (StoCron) because ifconfig is too dumb to get anything done 2005-05-05 [15:11:31] progress 2005-05-05 [15:11:44] the way BSDs extend ifconfig is actually making a pig fly 2005-05-05 [15:11:51] (StoCron) look at the way the BSDs configure GRE tunnels... you can't set anything interesting, except by "-link0" "-link1", etc. flags 2005-05-05 [15:11:54] i don't know, my cow-orkers usually tell me "keep things simple, [Solaris|HP-UX...] is too complicated 2005-05-05 [15:11:56] yes throw away stuff that has been used for 30 years and start from scratch 2005-05-05 [15:11:57] (StoCron) (at least on NetBSD, dunno about FreeBSD) 2005-05-05 [15:11:58] every 3 years. 2005-05-05 [15:12:01] the linux way. 2005-05-05 [15:12:01] I still wonder what all that has to do with the ap-wg jabber=>room relay 2005-05-05 [15:12:32] who's relying btw... gert, you? 2005-05-05 [15:12:36] relaying 2005-05-05 [15:12:50] (marcoh) eamonn 2005-05-05 [15:13:01] ok, thx 2005-05-05 [15:13:17] . o O(probably noone thinks there is any need on discussing the HD-ratio issue which is currently up) 2005-05-05 [15:13:30] (StoCron) so what do you think about HD-ratio values? 2005-05-05 [15:13:52] (UncleH) change it to 0.94 and see how it flies seems like the safe thing to do for now 2005-05-05 [15:14:07] (KnockandO) Wow I understood about half of this. 2005-05-05 [15:14:16] he,he 2005-05-05 [15:14:23] (StoCron) so start listening instead of IRCing all the time :-) 2005-05-05 [15:14:24] I do agree that 0.94 might be saner 2005-05-05 [15:14:28] (KnockandO) I did :) 2005-05-05 [15:14:36] (KnockandO) (got better towards the end) 2005-05-05 [15:14:38] having recently worked on an addressing plan for a /21 :-) 2005-05-05 [15:14:48] (StoCron) dr: can I quote that? 2005-05-05 [15:15:01] i read through it some time ago, it somehow makes sense, but i didn't really think about the issue anymore 2005-05-05 [15:15:13] if you're bored read my rant that I wrote after geoff's talk at a ripe meeing 2 years ago: http://www.bgpexpert.com/article.php?id=79 2005-05-05 [15:15:14] (marcoh) dr: v4 or v6 ? 2005-05-05 [15:15:22] marcoh: v6 2005-05-05 [15:17:04] (KnockandO) v7 2005-05-05 [15:17:14] hm, don't really get why we would need that for ipv4 thought, right 2005-05-05 [15:17:39] (StoCron) slz: same problem for large networks, if you have a /12 of DSL customers, 80% usage is hard 2005-05-05 [15:17:52] why? 2005-05-05 [15:18:21] hm, isn't that a dynamic pool in IPv4 anyways, so it's full from the beginning? :-) 2005-05-05 [15:18:23] thats why policy is approximate 2005-05-05 [15:18:28] currently 2005-05-05 [15:19:10] the telia guy is very right 2005-05-05 [15:19:16] I mean, it's not a real suggestion, but actually we should think about making things easier on IPv6 than on IPv4 so people start migrating :-> 2005-05-05 [15:19:20] (StoCron) dr: he's very confused 2005-05-05 [15:19:22] hehe, the NTT roadtrip ;> 2005-05-05 [15:19:23] (KnockandO) I believe, 80% is almost impossible. 2005-05-05 [15:19:33] (KnockandO) There's a lot of issues in a real-world network. 2005-05-05 [15:19:34] why would 80% be impossible? 2005-05-05 [15:19:37] (StoCron) slz: think "multiple locations and internal aggregations" 2005-05-05 [15:19:51] (StoCron) and not all is dynamic IPs - static assignments happens 2005-05-05 [15:19:53] (KnockandO) Think aggregation, think uneven location distribution and size, think customer classes, think routing protocols. 2005-05-05 [15:19:59] why is it hard? if i reaquest a /12, i guess i make the request already fill up >=80% in anyway 2005-05-05 [15:19:59] if you have a /20 that's 250 million /48s 2005-05-05 [15:20:03] (KnockandO) Think mobile customers :) 2005-05-05 [15:20:37] guess we're back on the "small lie" problem 2005-05-05 [15:20:44] (KnockandO) And of course you want to have room for growth without a new roll-out. You simply want to be able to upscale. 2005-05-05 [15:20:45] but you only have to assign some 6 million 2005-05-05 [15:20:52] sounds like the easiest thing in the world. 2005-05-05 [15:21:00] (for a 80% hd ratio) 2005-05-05 [15:21:30] (KnockandO) You have to route them, and certainly 6 million routes in your IGP looks kind of sick. 2005-05-05 [15:21:54] why would they be in your igp? 2005-05-05 [15:22:03] are your v4 customer /32s in your igp? 2005-05-05 [15:22:36] (StoCron) iljitsch: whatever protocol you use for internal routing - you don't want 6 million routes in there 2005-05-05 [15:22:37] (KnockandO) When we had customers, the aggregates were. And there was always a big waste of addresses due to aggregation. 2005-05-05 [15:22:37] StoCron: I'm referring to his point that "ease of ACL fomulation" _is_ an important thing to think about when designing addressing plans 2005-05-05 [15:22:47] * Djinh has joined the channel. 2005-05-05 [15:23:06] (StoCron) dr: sure, but "earmark a /40 for that and have all addresses that all your infrastructure will EVER need - and then setup filters for that" 2005-05-05 [15:23:06] like I said: we aren't doing this in v4 so why would this be an issue in v6???? 2005-05-05 [15:23:27] (StoCron) iljitsch: because there's more of it? There are MANY /48 inside a /20 2005-05-05 [15:23:36] yup, "enoigh for anyone"? :) 2005-05-05 [15:23:40] (KnockandO) iljitsch: Sure that the big guys (with millions of customers) don't? 2005-05-05 [15:24:29] someone here is very stupid... no idea who. but I don't get why we're even TALKING about customer /48s in the igp. 2005-05-05 [15:24:53] You never need to waste more than 1 bit per aggregation level. 2005-05-05 [15:24:59] (KnockandO) We are, if you want good usage percentage. 2005-05-05 [15:25:02] I don't know, but i know networks who have /32 (IPv4) in their IGP... 2005-05-05 [15:25:13] (KnockandO) Otherwise you're always wasting, and these numbers get very big in v6, AFAICS 2005-05-05 [15:25:16] (StoCron) can someone please explain to ijitsch how real-world networks work? growing over time, and so? 2005-05-05 [15:25:38] (marcoh) slz: ack, I need to support a certain level of roaming 2005-05-05 [15:25:51] I've made all these mistakes myself a long time ago. The thing is some poeople learn from that. 2005-05-05 [15:25:51] (marcoh) (across AS-borders even) 2005-05-05 [15:26:06] (KnockandO) Alright, you have a city with 9000 available ports, another with 17000 for customers, one with 2^8+1 ports... 2005-05-05 [15:26:10] how does this roaming work? 2005-05-05 [15:26:11] I can even tell that in the Siemens Corporate Network, we have /32 in the BGP tables soo.... :-) (ok but that's corporate network, different laws) 2005-05-05 [15:26:32] (StoCron) no matter what you put in which routing protocols - you NEED aggregation (and a number of levels), and each aggregation level WILL lose some fraction of the address space 2005-05-05 [15:26:39] maroch: yep, roaming issues mostly 2005-05-05 [15:26:41] (KnockandO) Unfortunately, real life doesn't aligh on powers of two. 2005-05-05 [15:26:52] (marcoh) iljitsch: simple, people can dialin in the UK with a NL account and viceversa...means we need to announce /32's when they do across our private peering 2005-05-05 [15:26:52] --- damien has left 2005-05-05 [15:26:53] that's why you lose a bit. 2005-05-05 [15:27:16] (KnockandO) And actually, 75% usage (meaning 25% loss) per level is gorgeous usage 2005-05-05 [15:27:17] with a static ipv4 address? cool. 2005-05-05 [15:27:27] (StoCron) one bit is not sufficient if you need to plan for some growth (and "1 bit" can mean "55% efficiency per level", which is pretty bad) 2005-05-05 [15:27:51] (marcoh) Knockland: you should put that on a t-shirt (the powers of 2 thingie) 2005-05-05 [15:28:00] planning for growth makes for wasted resources. grow first, make a new plan when you're done. 2005-05-05 [15:28:03] (KnockandO) marco: Just thought about it :) 2005-05-05 [15:28:15] (marcoh) knock: next meeting ? 2005-05-05 [15:28:23] (KnockandO) Let's see what T-Shirt printers can do :) 2005-05-05 [15:28:31] (marcoh) have denic sponsor it 2005-05-05 [15:28:42] right, DENIC seems to have too much money anyways 2005-05-05 [15:28:46] (KnockandO) :) Sometimes they are nerdy enough... 2005-05-05 [15:28:59] (KnockandO) slz: We do not have any money. It all belongs to our members... 2005-05-05 [15:29:06] (marcoh) point them to the netnod ones... 2005-05-05 [15:29:08] ko: ok, give me some :) 2005-05-05 [15:29:12] (KnockandO) Stocron: The problem is more the number of levels you need. 2005-05-05 [15:29:17] --- dr has left 2005-05-05 [15:29:17] (KnockandO) slz: Members? 2005-05-05 [15:29:22] <- member 2005-05-05 [15:29:35] (KnockandO) Oh, then you'll be returned some in December. Like every year.. 2005-05-05 [15:30:05] that would be fine, so we can pay the RIPE membership then . o O(just to get back on track here) 2005-05-05 [15:30:20] (KnockandO) .Ah right, we're in a RIPE channel here... 2005-05-05 [15:30:20] --- ferenc has left 2005-05-05 [15:30:37] --- fil has left 2005-05-05 [15:30:43] * JNC quit IRC altogether 2005-05-05 [15:30:46] (KnockandO) How many levels do you actually need? I would vote for 3 levels of aggregation in nationwide networks, 5 or 6 for world-wide. Reasonable? 2005-05-05 [15:31:06] nation == luxumburg or nation == china? 2005-05-05 [15:31:22] (marcoh) hmmm, coffee & nicotine 2005-05-05 [15:31:52] (UncleH) tea and a wee as paul described it so aptly yesterday 2005-05-05 [15:31:54] I don't think you can make any suggestions here, the engineering departmens of several ISPs usually have their own ideas in the end anyways 2005-05-05 [15:32:12] --- geoff has become available 2005-05-05 [15:33:12] (KnockandO) ALright, coffee 2005-05-05 [15:34:03] --- geoff has left 2005-05-05 [15:34:08] hm NCC Services wg next 2005-05-05 [15:34:42] i think i'll miss the ipv6-wg tomorrow :-( bad timing 2005-05-05 [15:35:11] i hope there will be recordings of the webcasts again as usual? 2005-05-05 [15:35:22] --- EamonnMcGuinness has left 2005-05-05 [15:35:26] --- iljitsch has left 2005-05-05 [15:37:30] (jluk) slz: I assume there will be. I can catch up on the talks that happen before I get out of bed then :) 2005-05-05 [15:40:24] * andy quit IRC altogether 2005-05-05 [15:43:09] * [rvs] quit IRC altogether 2005-05-05 [15:44:37] * geertn quit IRC altogether 2005-05-05 [15:45:02] * jluk quit IRC altogether 2005-05-05 [15:47:36] --- slz has left 2005-05-05 [15:53:26] --- jhma has left 2005-05-05 [15:59:53] * JNC has joined the channel. 2005-05-05 [16:01:51] --- Fearghas has left: Disconnected 2005-05-05 [16:03:55] --- rhe has left 2005-05-05 [16:05:20] * StoCron has left the channel. 2005-05-05 [16:25:38] * JNC quit IRC altogether 2005-05-05 [16:26:12] * Djinh has left the channel. 2005-05-05 [16:49:03] * prt quit IRC altogether 2005-05-05 [17:08:37] * KnockandO has left the channel. 2005-05-05 [17:11:08] * geertn has joined the channel. 2005-05-05 [17:23:25] * uk quit IRC altogether 2005-05-05 [17:33:41] * geertn quit IRC altogether