[00:23] *** EhtyarWRK has quit IRC (Quit: Never look down on someone unless you're helping them up.) [00:26] Huh? I have higher latency to ARP than maxamillion did, and VNC is fine for me. I can't see that he's going to be happy anywere, especially since 82ms of his 116ms ping times were internal to his WiMAX provider. [00:29] same here, i'm always 170-180ms (UK) and VNC works fine [00:37] Woah you get that good from UK? [00:37] 210 to Australia... It's improved actually. [00:53] 210 seems very high [00:53] it should be more like 170 for australia? [00:53] DDevine: yeah, 170-180 isn't too bad (about 15ms of that is my broadband connection) [00:54] http://traceroute.optusnet.com.au/?args=www.arpnetworks.com [00:54] dsl is about 10 msec latency on top of that i imagine [00:57] actually 3g could give you 210 too [01:08] mercutio: Could be my wifi. [01:12] DDevine: I'm also in the UK and get 140ms on IPv6 and 170ms on IPv4 to ARP, 15ms of each being my DSL [01:12] So bob^^'s latency seems about right from here [01:15] my latency from work (I work for an ISP) is only 150msec [01:15] (that's direct from the core of our network though) [01:16] and i get 154ms from home on v6 (via an HE tunnel) [01:19] bob^^: Same here. 150ms on v4 from a machine in THN, 130ms on v6. (I also work for an ISP :) [01:22] :D [01:23] 145ms out of our network in THE on v6 :( [01:23] interesting, the vast majority of that 145ms occurs inside HE's network [01:24] are you a linx member plett? :) [01:25] bob^^: We are indeed. AS20712 Andrews & Arnold [01:25] Who are you? Do we peer with you? :) [01:25] i think we do, yes :) [01:25] AS25178, Keycom PLC [01:25] Yep, we have v4 and v6 sessions at LINX :) [01:26] we do indeed, small world :) [01:26] are you going to the next linx meeting? [01:26] About a month's time? Windsor? [01:26] yup that's the one [01:27] I think so, yes. I've registered myself, but we generally don't know who's going until closer to the time [01:27] i'm down to go too - fingers crossed there's enough of us left in the office :) [01:28] We're based in Bracknell, which is a 20 min train ride from Windsor [01:28] ahh not bad at all - we're in Stafford, though I live in Stratford on Avon [01:29] bit further for the guys i work with, not so far for me :) [01:31] if you end up going give me a shout - i'll come say hi [01:31] Will do :) [01:32] hang on - was it you guys who had 'the internet' at one linx meeting a while ago? [01:34] bob^^: Yeah, that would be us :) [01:34] :) [01:34] More specifically that would be our MD, Adrian [01:34] yeah, that's right - he did a talk on v6, was interesting [01:34] more specifically about v6 capable CPEs iirc [01:35] Yes, that's one of our common rants. Everything else in the chain from an end user to (say) facebook is IPv6 capable. It's just their DSL router that isn't. [01:35] yeah, it's quite frustrating [01:36] more frustrating how few ISPs can actually deliver v6 to the house [01:36] i'm with Be, no sign of v6 yet at all annoyingly [01:36] They are starting to be produced though. We have a £100+ "Billion 7800N" which works fine with IPv6, but at that price it's not one we can just give away to every new customer [01:37] i've heard zyxel have quite a bit of v6 capable stuff now including a few home routers [01:38] They like to say that they have. The difficulty we have getting hold of them says otherwise. [01:38] bob: zyxel are still around? [01:38] i thought they made weird modems [01:38] yeah, zyxel make all sorts of stuff [01:38] weird dialup modems that is [01:38] i never see anything zyxel around these days [01:38] plett: ahh, not tried actually getting one :) [01:39] nor us robotics [01:39] US Robotics got bought up by 3com [01:39] years ago [01:39] ahh [01:39] :) [01:39] that's a name i've not heard for a long time [01:39] * bob^^ digs out his courier [01:39] v.everything [01:39] :) [01:40] mercutio: i just bought a zyxel nas for home (their kit isn't brilliant but it's alright for home imho) [01:40] i've never seen a home nas that was any good [01:40] this one has two bays and does raid etc - only had it a couple of days but so far so good [01:41] not gonna set the world alight with it's feature set or performance, but good enough for me :) [01:41] bob^^: We have a "ZyXEL P-2612HNU-F1" here in the office for testing, which they sent us this week. It came with an european plug, as they haven't made a UK model yet, required a beta firmware flashing to it to make it even claim to do v6, and then doesn't work. [01:41] oh, that's promising then! [01:43] speaking of ipv6 [01:43] whatever happened to the internet having larger muts [01:43] mtus [01:43] what with ipv6 having higher overhead and all [01:43] surely it's time for it to get bumped up a bit [01:43] *** LT has joined #arpnetworks [01:43] It's stupid though, as all the manufacturers of cheap routers use the same chipsets, and use the chipset manufacturer's canned "build me a firmware" GUI wizard thingy. And that wizard has supported v6 for about 3 years now, but nobody ever ticks the box to enable it [01:44] hah, i didn't know that [01:44] plett: real [01:44] maybe they don't do it in hardware [01:44] larger address family != larger mtu [01:44] do cheap adsl modems do checksum offloading etc? [01:44] toddf: i know [01:44] our customers are all connected over ethernet so (apart from a few on DSL) thankfully we don't have to worry too much :) [01:44] but because it's using up more overhead [01:44] and that it's not in serious production use [01:45] surely it's about time for mtus to rise in size [01:45] Yeah. If you've ever wondered why the user interface on cheap routers all look very similar, it's because they _are_ all the same. The "manufacturer" just copies the reference hardware design, cutting as many corners as they can get away with, and then uploads their logo and brandname into the firmware wizard, and ships it :) [01:45] i dunno the few modems i've used the gui on have all been radically different [01:45] dlink, linksys, tp-link [01:45] mercutio: I think you oversimplify the task at getting everybody to participate in the larger mtu crowd [01:45] mercutio: jumbo frames would be the only option, and not everything supports them [01:46] (assuming ethernet is the technology being used, obv) [01:46] toddf: well i'm not saying it's easy, but it's not easy to get ipv6 going either [01:46] (and in our case everything core is based on ethernet) [01:46] mercutio: Cheap DSL modems/routers do everything in software. These days they are a single IC with combined CPU, ethernet and DSL, but there's still no offloading etc [01:46] mercutio: there are larger mtus, they're called jumbograms, and if all os's behaved sanely, you'd set max mtu of your hardware, and 'discover' (path mtu) the max mtu available per remote host, end of story [01:46] mercutio: IPv6 is pleanty easy, I've been doing it for 10+ years [01:47] toddf: well lots of dsl has mtu < 1500 [01:47] mss clamping kind of helps [01:47] toddf: but it's not really serious yet [01:47] mercutio: PPPoE has a default of 1492, unless you have modern enough kit to support RFC4638 [01:47] discovering max mtu is the key to anything greather than and less than standard mtu size [01:47] plett: even in pppoa 1492 is pretty common [01:47] ob [01:47] on [01:47] and then there's ipsec etc [01:47] mpls [01:48] people running mpls on 1500 byte networks [01:48] etc [01:48] mercutio: you can say its not serious, but I use it daily for 90% of my traffic, so however you define serious is up to you [01:48] toddf: 90%?! [01:48] what do you do over ipv6? [01:48] mercutio: Really? My only experience with DSL is in .uk , but here all PPPoA is clean 1500 [01:48] plett: here pppoa is common [01:48] mercutio: everything. dns64 / nat64 / even tunnel afs over v6 back to my fileservers, smtp, www, dns, imap, ping, ... etc etc [01:49] but the telecoms provider was routing over pppoe afterwards [01:49] you'd think they'd just raise the mtu on the ethernet segments [01:49] but like when i look at traffic on the net in general [01:49] that's not particularly straightforward [01:49] lots of people have mtus of 160 etc [01:49] you'd have to ensure that everyone you connected to ran the same jumboframe size (again, assuming ethernet) [01:50] and that all your kit supported it (not all kit supports jumbo by a long way) [01:50] bob: yeah a bit doesn't [01:50] maybe it's hopeless [01:50] but it seems weird to have 10 gigabit ethernet etc around, and 1500 byte mtus [01:50] and even 100 mbit ethernet could handle bigger mtus [01:51] increasing mtu would probably be many times more difficult than trying to implement v6 end-to-end :) [01:51] most gigabit stuff handles jumboframes [01:51] yup you can do jumbo on 100mbit on some kit (extreme networks kit has done jumbo for years) [01:51] mercutio: Here, the telco is most often BT. They do PPPoA on the tails from the EU to the exchange, and then trunk it over PPPoE in their backhaul network. They use a 1600 byte MTU for that, which easily fits the 1500 MTU + PPPoE header [01:51] plett: well yeh that's the sensible way :) [01:52] to my mind the biggest issue with adsl these days is the upload speed though [01:52] and annex-m isn't supported here [01:52] that's a problem with more than just adsl [01:52] mercutio: There are lots of other ways that DSL is deployed, and most of them are less sensible :) [01:52] you can congest it really easily [01:53] my line is annex m'd but i still only get 1.2mbit up [01:53] bob: real [01:53] bob: i get 1.2 mbit up without annex m [01:53] exactly [01:53] it's not as good as you'd think [01:53] what do you get without annex m? [01:53] i actually need to re-route some cables here, only recently moved in to this flat and the phone point is miles away from the modem (no power near the phone point) [01:53] just under 1mbit [01:53] mercutio: bob^^ said he is on Be at home. They don't use PPP at all. The ethernet coming out of the DSL modem goes straight onto your LAN, and the default gateway for your desktop machine is on the other end of the cable in the exchange. [01:53] bob: i have the same problem [01:54] so i'm using extension cable [01:54] indeed plett [01:54] it kills 1mbit off my down sync rate [01:54] plett: oh weird [01:54] Be has been pretty good for me so far - we use them sometimes (buying through Cerberus) for backup links in work [01:54] so it's like bridged plett? [01:54] like cable etc usually is [01:55] i did actually consider going with AA plett :) [01:55] bob^^: Good good :) [01:55] for the amount my line is used it was a little pricey though :( [01:55] (bandwidth hungry housemate0 [01:55] bob: you "needed" it though? [01:56] ? [01:56] oh it's used a lot you mean? [01:56] i thoughtyou meant it was hardly used [01:56] oh no, my broadband is pretty heavily used [01:56] i use a lot of upload (i back everything up to a colo box at work, etc) [01:57] mercutio: I don't know much about how cable is set up. I didn't think it bridged the EU's LAN out to the cable head-end, at least not here in .uk [01:57] plett: here they used to have one huge big arp domain [01:57] i used to have cable in my old flat [01:57] enough to overload most routers [01:57] well most of the shit routers that people tend to use [01:57] the ones that top out < 30 mbit [01:58] there's no cable reselling here [01:58] but ther eis for dsl [01:58] so cable is expensive [01:58] but better technology [01:58] ilke i think they upgraded to docsis 3 [01:58] cable here is priced similarly to dsl on average but we only have one cable provider and no resellers [01:59] i don't know why there's so many people keen on ftth [01:59] when docsis 3 is good [01:59] and has existing cable [01:59] cheaper etc [01:59] not much of the UK has cable [01:59] bob: why are you on dsl then? [01:59] ahh ok [01:59] there's cable in the town i live in [01:59] but not down my street [01:59] when i had a little look on the web at cable in UK it looked good [01:59] there was virginmedia? [01:59] that's always the problem with cable [01:59] and like 50 mbit [01:59] calbe [01:59] yeah, that's it [01:59] yup [02:00] mercutio: What makes you think cable is a better technology? [02:00] it's a good service actually - i used to have the 20mbit cable product, worked great [02:00] plett: then what? [02:00] I assumed you meant better than DSL [02:00] oh cable is more reliable than dsl [02:00] like [02:00] most people have old shitty wiring [02:00] line faults etc are common [02:00] dsl routers are usuaally shit [02:01] cable has those issues too tbh [02:01] Ahh. Okay [02:01] and break reasonably often [02:01] yeah it doesn't seem to be as often [02:01] and it can support higher bandwidth [02:01] twice i had to have engineers out to 'fix' the coax between my house and the street cab [02:01] bob: oh real? [02:01] i've had cable twice [02:01] err three times [02:01] four times? [02:01] shit [02:01] but in 6 years total of DSL, had an engineer out once and he had to replace the line... because.......... [02:01] i've moved around a lot [02:01] (And this is a great story) [02:01] bob^^: To be fair, no copper in a cable network is going to be much older than 1960. I've seen phone lines that must be 100+ years old :) [02:01] anyway, the only problems were with the isp doing transparent proxying [02:02] and with the aforemented arp domain issues [02:02] there was a fracture somewhere in my line - on hot days the cable expanded and for some reason the fracture 'fractured more' and i lost sync [02:02] whereas i've had dsl in multiple places too [02:02] so during the summer there were afternoons where i'd get no service :) [02:02] and had problems with drop outs line noise, problems when it rains etc etc [02:02] plett: very true - and cable won't have the MK-style alumninium lines either ;) [02:03] ripping off the bell wire here can be a big improvement [02:03] and docsis 3 can do 100mbit+ [02:03] ftth can do anything at all though [02:03] ftth is expensive though [02:03] bob^^: Ironically, it's the older phone lines that are more reliable for DSL - over the years as copper prices have increased, the wires have got thinner and thinner :) [02:03] fttc i agree though i don't really see the point - surely easier to concentrate on ftth [02:03] it costs a lot of money to run the fibre etc [02:03] so if ther'es already cable [02:03] yeah plett, very true! [02:03] it seems pointless to run fibre [02:03] i could get 50mbit down 25mbit up with docsis 3 today. I just don't want to pay $384/mo for it. [02:04] some of our customers are on 50mbit down/50mbit up [02:04] here they're doing vdsl [02:04] which uhh [02:04] ...but they're delivered over straight ethernet so that's pretty easy and cheap to do :) [02:04] hasn't ahppened properly yet [02:04] mercutio: DOCSIS 3 can do 100Mb, until a second person in the same broadcast domain (typically several thousand houses) wants to use it at the same time. At that point you have to share your 100Mb :) [02:04] yeah, our FTTC is vdsl for the last mile [02:04] and there were plans for fibre i think, but people are concentrated on available bandwidth rather than performance/reliability/latency/international transit etc etc [02:05] plett: it does over 100Mb total though doesn't it? [02:05] cable here was 15mbit [02:05] it was always pretty good for those speeds nationally [02:05] international it was fucked [02:05] especially if used web [02:05] because it hit a transparent proxy with small window sizes [02:06] which tended to get evening peformance degredation etc etc [02:06] leading to the "bittorrent is fast but web is slow" dilemma. [02:06] which also happens on congested networks without shaping/qos [02:07] so peopla re like - can pull line rate with bittorrent - it must be the remote servers. [02:07] is transparent proxying used in the UK? [02:07] Depends on the ISP [02:08] In our case, we don't do any proxying, filtering or shaping. IP packets in == IP packets out [02:08] i reckon it actually makes sense for international stuff [02:08] it used to happen on cable here mercutio, but i think they've (mostly?) stopped now [02:08] plett: we're the same [02:08] much easier and fairer imho [02:08] Indeed [02:08] fairer? [02:08] you pay an ISP to transit packets from A to B [02:09] if they interfere with the packets in-flight that doesn't seem very fair [02:09] hmm [02:09] what do you think about explicit proxies? [02:09] i don't see the point these days [02:09] ihmo, there is no need for proxying today [02:09] because cdns are used more? [02:10] because transit/peering are so cheap and content is normally local through a CDN [02:10] heh [02:10] transit is expensive here [02:10] the things that really eat bandwidth aren't cachable anyway... eg youtube [02:10] i'm in new zealand [02:10] it's cheaper to just buy more connectivity than to build a proxy cluster [02:10] ahh, that would be an issue then :) [02:10] lt: google provide caches [02:11] actually that's true, they do indeed [02:11] bob: i think there can be higher performance, when it's done proeprly. [02:11] google will happily drop a cache node in to your network if you meet some criteria and don't mind giving them half a rack and some power [02:11] yaeh it's some amount of sustained traffic, depending on your country [02:11] i disagree, i'm really not a fan of proxying at all [02:11] lowest in south america i seem to recall [02:11] highest in US [02:12] well they brand it as a cache... but isn't it really just a cdn node? [02:12] LT: it's a cache [02:12] it forwards along to the closest peer [02:12] http://ggcadmin.google.com/ggc [02:12] unless it's already downloaded it before then sends it direct to user [02:12] blurry.... a cache that only works for google stuff, is kinda different to a traditional cache [02:12] it's a smart idea [02:13] I don't see it's much different from an akamai box [02:13] LT: it's pretty similar? [02:13] yeah, it's not really [02:13] It's more of a dynamic CDN [02:13] oh [02:13] it's the same as akamai [02:13] just for different content [02:13] however - if google+ catches on, it might be good for users and for ISPs [02:13] it really depends if you have peering to a google node directly or not [02:13] whether it's worthwhile [02:13] yeah [02:14] ilke don't google peer over linx for you guys? [02:14] yeah. it's sensible enough... but I got the impression you were talking about sticking all http through a proxy, which is a slightly different beast [02:14] We already peer with Google at LINX and LoNAP, so already have zero bandwidth costs for traffic to them [02:14] hehe [02:14] i can't remember if we peer with google or not [02:14] we peer with a GGC on MaNAP [02:14] (or whatever it's called today) [02:14] they have an open policy don't they? [02:14] bob^^: Edge-IX, I think [02:15] yes, we peer with google on linx now too [02:15] so yeah, our traffic to google is free too [02:15] that's the one, Edge-IX :) [02:15] that said [02:15] if you were starting to congest your peering link [02:15] And, for UK networks, Google server content from their Ireland datacentre, so it's relatively low latency too [02:15] you could get google cache [02:15] mercutio: you'd just get a private interconnect [02:16] bob: oh true [02:16] that's another way to go [02:16] it'd be nice if google was in nz :) [02:16] they must have a presence there? or in australia? [02:16] australia [02:16] nz<-> australia connectivity isn't amazingly cheap [02:17] it's the same monopolistic cable that runs to US [02:17] just different segment on it [02:17] hmm, it is interesting to see different peoples opinions [02:18] i've been working on an explicit proxy mesh system to accelerate web browsing [02:18] by routing to a proxy near the end destination [02:18] and keeping persistent connections open to the proxies [02:18] reduces latency etc etc [02:19] In the case of NZ, international traffic is always going to be expensive and high latency, just because of geography and the speed of light. The only way round that is to either serve content from local servers, or do caching [02:19] plett: yeah - but - i've found that NZ<->UK is extra shit [02:20] it goes via the US [02:20] mercutio: It mostly goes via USA? [02:20] Yeah [02:20] and so it's like 260 msec minimum [02:20] but [02:20] web sites won't send you more than like 4k of data in one round-trip-time [02:20] so it's like .. 4k.. 8k. . 16k.. [02:20] etc etc [02:20] assuming no packet loss [02:20] The other route would be via Asia, and taking an over-land route to europe [02:20] but on top of that in my testing, it seems some uk sites are slow etc too [02:20] plett: asia routing is /messy/ [02:21] Indeed [02:21] plett: that's the way planes fly though [02:21] so sometime in the distant future it may work that way [02:21] I'm not surprised that the BGP hop count is lower for traffic via USA than via Asia [02:21] in my testing, i've found that guardian.co.uk is faster on average than bbc.co.uk [02:21] but then i've tried curl frm a uk host to bbc.co.uk [02:21] and i find weird 200 msec delays etc [02:22] seemingly randomly [02:22] strange [02:22] even with a 2msec ping or something [02:22] oh is that not normal? [02:22] plett: There is a lot of countries in APAC... lots of hops. [02:22] mercutio: I've not seen that here [02:22] plett: things don't go in a direct line [02:22] oh, i only have one uk host [02:22] so maybe it's that host [02:23] but latency doesn't seem to spike [02:23] time curl --compressed http://www.bbc.co.uk/> /dev/null [02:23] like what's that say for you? [02:23] real 0m0.083s [02:23] user 0m0.004s [02:23] sys 0m0.000s [02:23] oh that's fast [02:24] right now i got 1 second, 113 msec, 130 msec, 60 msec, 131 msec, 130 msec [02:24] from nz it's way worse though [02:24] That's from my desktop in the office, which is gig-e or higher all the way to the bbc [02:24] 1.981, 1.920 [02:24] 1981 msec that is [02:25] for what, 25k of data [02:25] then you have all the images etc etc [02:25] From my DSL at home: [02:25] real 0m0.179s [02:25] user 0m0.008s [02:25] sys 0m0.008s [02:25] 0.016u 0.008s 0:00.18 5.5%288+1668k 0+0io 1pf+0w [02:25] (from work) [02:26] from my proxy is 990 msec, 726 msec [02:26] i think bbc has low ttl [02:26] on dns [02:26] from home dsl: [02:26] real 0m0.315s [02:26] user 0m0.000s [02:26] sys 0m0.020s [02:26] hmm [02:26] 315 msec is ok [02:26] (i'm using my connection atm to watch f1 free practise over iplayer too!) [02:27] 2 seconds isn't :) [02:27] wow your guys times look diff to me [02:27] curl --compressed http://www.bbc.co.uk/ > /dev/null 0.01s user 0.00s system 7% cpu 0.130 total [02:27] like mine just shows on one line [02:28] i suppose that's bash [02:28] my colo is freebsd and my desktop at home is ubuntu [02:29] probably just differences in 'time' [02:29] Differences in the time command, I would expect. My examples were both Ubuntu [02:29] hmm i get the same on linux and openbsd [02:29] this was freebsd: 0.016u 0.008s 0:00.18 5.5%288+1668k 0+0io 1pf+0w [02:29]