***: HighJinx has joined #arpnetworks
ivan-kan` has quit IRC (Remote host closed the connection)
ivan-kanis has joined #arpnetworks
mig5 has joined #arpnetworks
mig5 has left mercutio: does arp have a looking glass? dxtr: Wat? RandalSchwartz: a mirror? :) ***: heidar has quit IRC (Read error: Connection reset by peer)
heidar has joined #arpnetworks G: dxtr: you know, a queriable BGP server type thing so you can see how the internet appears from the prespective of different places on the internet RandalSchwartz: why would arp need one? G: RandalSchwartz: I guess it's just a 'nice to help' but personally I don't see the point in every provider doing it RandalSchwartz: Yeah - how many different views do you need? :)
you can also just traceroute yourself. I do that a lot. plett: I'm a new customer, and would have found one useful to see how a VPS at arp would route traffic to various places before I signed up RandalSchwartz: since it varies from day to day, how does that help? G: RandalSchwartz: there are useful aspects though, but as I said, I don't see the point in every provider doing it plett: BGP shouldn't really be that unstable RandalSchwartz: if you make purchase decisions based on a particuler view one afternoon, I think you're missing the picture. :0
plett - arp has a lot of redundant links
so yes - it does vary, with load, with what's up vs down, etc. G: RandalSchwartz: Looking Glasses seem to only be of any use if stuff is broken RandalSchwartz: and if stuff is broken, up_the_irons should be fixing it. not waiting for us to diagnose it remotely :) G: RandalSchwartz: I'm talking in a general sense RandalSchwartz: I generally do too. :) G: and who's to say the brokenness is up_the_irons' fault RandalSchwartz: I didn't say that plett: RandalSchwartz: Indeed. I have seen the lists of peers on arp's website. It's the details of how it would be routed in the normal situation of all transits and peerings being up that I would have been interested in RandalSchwartz: plett - again, depending on load. :)
there is some sort of ASN mapper out there. G: plett: why not traceroute from the other end? RandalSchwartz: ok - time to go get on a plane from Porto Alegre to Rio. -: RandalSchwartz waves off plett: G: I did. The routing is asymetric in some cases
RandalSchwartz: BGP doesn't include the load on the links when chosing a route for a prefix, unless arp are doing something non-standard
Routes to the UK (which is mostly what I'm interested in) improved quite a lot when arp added PacketExchange/mzima/GTT/whoever_they_are_today transit
A looking glass at ARP's end would let me see where it would go if the PacketExchange path were down, etc mercutio: plett: hah it's mzima routing that's borken for me
well i mean suboptimal
i was wondering how much it's preferred by and what my transit provider is doing
it used to be fine
oh actually it was only mzima foward path to the vps
reverse path is via trit.net
i dunno who trit.net is
mzima seems to be forward path going via paix.
it's easier to change forward-path routing than reverse-path routing though
https://www.trit.net/corp/
wow
web site coming soon plett: mercutio: From here in the UK, it's mzima in both directions and a nice short route. It helps that we peer with PacketExchange/mzima in London, and ARP has transit from them in LAX mercutio: plett: what's short?
like what latency?
i have a uk vps
hmm it goes via mzima 150msec plett: I was meaning hop count rather than latency, but I get 160ms from my ADSL at home mercutio: that's not too bad
i suppose
i'm in new zealand plett: Ironically, IPv6 is consistently 20ms lower - 140ms rather than 160 mercutio: my latency was around 126 msec not including dsl segment
now it's like 167 not including dsl segment
dsl is around 10 msec
plett: that happens soemtimes
with one provider that i used to use it's 133 msec frmo uk
anyway the difference between 169 msec and 126 msec is actually noticable plett: IPv6 being lower latency is mostly just down to "everyone using HE" mercutio: but i realise that it's probably from my transit provider having fucked routing now
hmm
he.net is good
in osome ways
not very reliable
but some of their routing is pretty good
although on some paths it can vary
like going via denver/chi or phoenix/dallas/atlanta plett: We only get v6 from them in London as they peer for free mercutio: i usd to think he.net was just a budget lame provider
oh yip
i've got he.net tunnel plett: I have no knowledge of what their v4 routes are like mercutio: to the uk they're pretty good
hmm 155msec frmo san jose
that's not that good
oh that's right
they're better from la to uk
then san jose
san jose goes via ashville/paris
los angles gos via new york/london
i dropped the la provider that used he.net
due to excessive packet loss though
don't think it was he.net's doing though
3. lonap.he.net 0.0% 8 10.2 2.8 1.3 10.2 3.1
4. 10gigabitethernet4-4.core1.nyc4.he.net 0.0% 8 70.6 72.4 69.3 79.4 3.9
5. 10gigabitethernet8-3.core1.chi1.he.net 0.0% 8 86.7 87.2 86.6 88.9 1.0
6. 10gigabitethernet3-2.core1.den1.he.net 0.0% 8 119.6 111.8 110.3 119.6 3.2
7. 10gigabitethernet1-4.core1.lax2.he.net 0.0% 8 132.7 134.2 132.5 142.0 3.3
but athat's what their london to los angeles routing is like
who was it before mzima? plett: Originally it was all PacketExchange, then they created the mzima name for transit mercutio: nah i meant that arp was using plett: Oh, It used two or three different transit providers to them from here
I don't know about the other direction, as I didn't have anything at arp at that point mercutio: really packet loss makes more diff than latency
like 5% packet loss is worse than 40msec more latency plett: Looking at the BGP tables from the London end, every transit feed we have has picked packetexchange as its best route to arp mercutio: heh plett: And for v6, everyone (apart from Level3 of course) uses HE to get to arp mercutio: what do your transit tables say for best route to 202.49.71.59?
(my IP in new zealand)
it mostly goes via cogent from uk i think to me plett: Only if the UK isp is using cogent, I would assume
No short route to 202.49.71.59. Best route from my view in the UK is via Level3 - 3356 4648 4610 17746 9559 mercutio: i dunno it may be peered
hmm
i'm guessing cogent is one of those asnumbers
oh hey it's not plett: No, Cogent is 174 mercutio: do you get less than 300 msec ping? plett: That's Level3 -> Netgate -> Odyssey -> Orcon -> Plain Communications mercutio: yip plett: None of those names mean anything to me, they all look .nz to me :) mercutio: heh plett: 295ms mercutio: netgate is big provider here
odyssey is pretty much transit of orcon
orcon is upstream
plain communications is a small isp i work at
295 msec yeh.. plett: I also work for a small ISP, but in the UK :) mercutio: that's where latency gets annoying :)
ahh ok
do you do many diff things ?/
it's weird with big companies how people like really specialise plett: The usual. DSL, ethernet connections, voip, etc
Most of our income is DSL mercutio: nah i meant do you
not the isp plett: Ahh, yeah. Anything technical mercutio: ahh yip :)
that's basically my role heh plett: Most small companies are the same - everyone does everything mercutio: shit we don't make shit all off dsl.
none of the providers here make much off dsl
ethernet, voip etc have more profit plett: Ethernet has bigger profit margins, but you sell it to far fewer people mercutio: true
but like dsl
there is so small margins
i dunno, like here, you pay a lot to one company who owns dsl equipment
that connects to the users etc
so
that basically eats into things
and also bandwidth is expensive here
that said all the isp's here have data caps plett: Here it's British Telecom who owns the majority of the copper to peoples houses and DSL DSLAMs etc. We buy end user DSL tails from them mercutio: yeh,
so probably similar deal.
i have a friend in the uk
in cambridge
he was using dsl
his dsl was really shit
i think it had interleaving on
but like he got poor sync speeds.
then he moved, and he got slightly better syn cspeeds.
speeds
but it seems like exchanges are often far away there?
like he was at 30 or 40 db attenuation
something really high
but like here they're doing cabinets everywehre to give people faster sync rates
so most people sync at between around 8 to 20 mbit plett: Cambridge is a bit of an anomaly. It has lots of high tech business and a university etc, so has lots of high tech fibre connections. But apart from that, it's a small towm with copper that has been there for 60+ years mercutio: i sync around 2100/1260
oh
so that's why his dsl sucks? plett: I know other people in cambridge, and they all say the same mercutio: he said that he wouldn't get virgin cable cos they wouldn't stop pestering him when he told them that the person that had cable connected when he moved in has left.
and were impossible to deal with
but i think virgin did 50mbit cable or something
err 2100 i meant 21500 plett: They're rolling out fibre-to-the-cabinet here too. I have it here at home, and get about 20Mb down and 10Mb up mercutio: vdsl? plett: Yeah. VDSL2 mercutio: there's vdsl2 equipment here
like i'll be on a vdsl2 port
but no-one is selling it yet plett: Maximum sync rate is 40M down and 10 up mercutio: oh and atm in t he trails you have to have interleaving on
so it's high ping
i used to live in a different city
and had like 20 msec higher ping
but like the diff between 10 and 30 msec ping
is actually quite noticable to me plett: Fibre-to-the-premises is on the roadmap here too. That will give 100M downstream mercutio: heh
thing is
web browsing doesn't go anywhere near 100 mbit plett: It's all the streaming video which is the killer mercutio: not here :)
yet.
well i suppose there's youtube
but like
that doesn't do anywhere near 100 mbit plett: We're a UK ISP, and Wimbledon tennis tournament is on this week - lunchtime data usage is sky high on our business customers mercutio: heh
is there a cdn for it? plett: The BBC stream it all live. They use Akamai and Highwinds as the CDN for it mercutio: which you'll have plenty of capacity to?
hmm, highwinds connectivity to nz isn't great
if its' the one i think it is plett: Oh yes. gig-e ports to Akamai. Highwinds refused to peer with us as we're too small mercutio: that's weird?
i thought cdn's would have open peering :) plett: The capacity problems come on the DSL backhaul links to BT
You'd think so, wouldn't you. Maybe that's why they're losing out to Akamai ;) mercutio: oh right
that's what was happening here
then they fixed it
well kind of
you have to change the way you're doing things
to connect
akamai is really slow i reckon
like with uncached content
well here it is at least
i honestly though they'd use more smarts plett: Peering is very big in the UK. Over half our traffic goes via peering rather than paid transit mercutio: wow!
about 10% of traffic is peered here?
i can't find a highwinds test download plett: We have huge peering LANs that have hundreds of members on them in the UK. I understand it's done differently in other parts of the world, with smaller per-building or per-provider public peering LANs, and lots of direct cables for private peering mercutio: anyway
hwcdn.net goes to LA
here like
the biggest providers don't peer
but the smaller ones do
but like
there's no google or anything here
so like there's not a lot of content peering
and like bittorrent etc
goes all over the place plett: Well, the big content providers or end-user eyeball networks peer here. But the big transit providers fairly obviously refuse to. mercutio: but usually not to peers
much
yeh
but like BT wouldn't peer right?
and virginmedia wouldn't?
and you were saying highwinds don't with small providers plett: Virgin don't, no mercutio: so it's interesting you can still get 50% peering traffic
i suppose youtube
accounts for a bit? plett: Yeah, we peer with google and facebook, which account for lots of traffic mercutio: cool
i suppose it is harder here
i've been trying to develop a system to improve performance over high latency connections atm plett: With .nz being thousands of miles from anywhere else? mercutio: err over the internet i mean
cos like international web browsing sucks from here
i've been like proxying with persistent connections
using geo ip lookups
to send to close proxies
i've kind of figured out that it wroks
works plett: I can't see an easy way to fix that, apart from dragging .nz somewhere else, or laying lots more undersea fibre mercutio: but now i want to try doing tcp acceleration
mm,
well my proxy system does google's initcwnd thing
like with sending more pcakets initially
which helps a lot plett: Ahh, window size stuff mercutio: but also like lots of sites don't have window sizes that get big etc
yeah
but even being closer to a destination
if a site has a small amount of packet loss
and you're close
resends have to go less distance
so as long as you don't have packet loss to a site that's near it
if the site near it has a big of packet loss to destination it doesn't hurt nearly as much.
but like, i suppose i'm a bit of a geek, i just want to see how fast i can actually get web browsing to get :)
which means having good peering etc.
but i can't think of any cheap way to do that
in my tests i've noticed that you have to be within about 20 msec from destination to get good speeds.
but like the internet is a huge place.. you can't really be 20 msec from everywhere
like norway is over 20 msec frmo london
but how many web sites other than opera are in norway?
actually looking at a map, i think norway could be 20 msec from london if there was a direct cable
i assume there probably isn't plett: I've got to run
mercutio: Nice talking to you. ***: dxtr has quit IRC (Ping timeout: 250 seconds)
dxtr has joined #arpnetworks
ivan-kanis has quit IRC (Remote host closed the connection)
ivan-kanis has joined #arpnetworks
ivan-kanis has quit IRC (Remote host closed the connection)
ivan-kanis has joined #arpnetworks
ivan-kanis has quit IRC (Remote host closed the connection)
ivan-kanis has joined #arpnetworks
ivan-kanis has quit IRC (Remote host closed the connection)
ivan-kanis has joined #arpnetworks
ivan-kanis has quit IRC (Remote host closed the connection)
ivan-kanis has joined #arpnetworks
ivan-kanis has quit IRC (Remote host closed the connection)
ivan-kanis has joined #arpnetworks
ivan-kanis has quit IRC (Remote host closed the connection)
ivan-kanis has joined #arpnetworks
ivan-kanis has quit IRC (Remote host closed the connection)
ivan-kanis has joined #arpnetworks
ivan-kanis has quit IRC (Remote host closed the connection)
ivan-kanis has joined #arpnetworks
ivan-kanis has quit IRC (Remote host closed the connection)
ivan-kanis has joined #arpnetworks
ivan-kanis has quit IRC (Remote host closed the connection)
ivan-kanis has joined #arpnetworks
homosaur has joined #arpnetworks
homosaur_ has joined #arpnetworks
homosaur has quit IRC (Ping timeout: 258 seconds)
homosaur_ is now known as homosaur
ivan-kanis has quit IRC (Remote host closed the connection)
EhtyarWRK has joined #arpnetworks
jebba has joined #arpnetworks
heidar has quit IRC (Read error: Connection reset by peer)
heidar has joined #arpnetworks
homosaur has quit IRC (Quit: pocketful of goat cheese, ready to party)