that google mtu thing gets stranger, even ip's of google caches have strangely small packets looks like 1454 mtu routing weirdness http://paste.unixcube.org/k/f7eaf savvis suck slashdot is using savvis and has huge pings too and doesn't show anything weird in traceroute from arp other than huge ping so i imagine savvis just have messed up peering with qwest.net as shown in your trace and peer with them in london funny because Qwest and Savvis are both CenturyLink i didn't know that just played with the PCCW and Savvis looking glasses looks like they peer in LAX oh oh that wasn't your normal connection that was from arp *to arp it's 160 msec to slashdot.org which is hosted in chicago (from arp) yea probably by London yeah but it looks like it's los angeles to chicago :/ forward path from ARP probably is the return path is screwed up ping jumps from 1 msec average to 156 msec average in single hop yeah it has been for ages but it's savvis -> qest -> pccw -> trit the qwest -> pccw is in London... does the looking glass show any paths? I didn't look at bgp the savvis -> qest is in london too isn't it? oh maybe not it is cr2-ten-0-7-4-0.londonuk1.savvis.net ... 63-235-40-189.dia.static.qwest.net i don't think level3, and ntt have amazingly close connections with savvis i think cogent/comcast/etc are closer that kind of explains it the funny part is that pccw and savvis peer in LA that would mean savvis carrying traffic to la I tried the Savvis lg traceroute from LA... same thing oh i'm just trying that now heh LA -> London -> LA :P but tracing to 63.218.212.14 (hop 9 in the trace to ARP) goes directly via LA uhh NTT have direct connection like forward path to slashdot is AS2914 AS3561 so it should be going via NTT instead of savvis? yeah but i'm struglging to use bgp ipv4 unicast is what i want? it's not coming up with anything idk anything about bgp i know i hate savvis's looking glass :) telnet://route-server.savvis.net damn it's not working that would have been really nice.. yes most of the looking glasses are really slow can you explain why when tracing directly to 63.218.212.14 from the Savvis lg, the path goes via lax (1.419 ms) i dunno what that ip is but when you trace arpnetworks.com, the hop 63.218.212.14 has a latecny of oh it's pccw i just did traceroute from pccw looking glass and saw that ip it's 0 msec from pccw to arp we kin dof need to see what routes they see 143.880 ms hmm ok what it's not seeing arp at all? atlanta quickly finds no routes los angeles slowly doesn't seem to do anything % Network not in table :/ https://groups.google.com/forum/#!topic/nanog-/MLWLu1NQZpI so it's been broken for years ah ok but yeah i assume savvis don't send to ntt because pccw are cheaper than ntt to send to? maybe pccw are customer it's still super strange that it would be going via london imo yeah good luck getting it fixed :) the Savvis NOC mailbox is full... I was trying to ask them about their looking glass / route server haha well if it was broken in 2011 i imagine they don't care much it's always fun to listen to NSPs tell me stright up that their crap is broken and that they're not going to fix it https://www.google.com/intl/en/ipv6/statistics.html#tab=ipv6-adoption wonder why the green line is so squiggly the interesting thing is on the very last datapoint, "6to4/Teredo" is 0.00% rather than the 0.01 and 0.02 it's been throughout 2013 and 2014 I wonder if that's related to the broken PTMUD problem yeah that is strange it maybe that they're trying stuff on/off a lot to measure performance how do ip address allocations work when you colocate? For instance if I have an ASA firewall and get assigned say a /29. Does the WAN interface of the firewall have to have a different /30 assigned? Allowing the entire /29 to be used for hosts, VMs etc? there's more than one way people do stuff you can get a /29 routed to you over your normal ip address, or you can have just the /29 if you have just the /29 you'll usually have 5 ip addresses to play with err that was a generic response with arp dedicated server i have a /29 and just the /29 I guess I am curious how that setup works when you have multiple devices, rather than a physical server cable comes in > connects to firewall > one port for physical web server, another for physical mail server, another for physical DNS server, all running on their own public IP address. I would thing they would point to the firewall for their default gateway s/thing/think I would think they would point to the firewall for their default gateway but what happens on the uplink interface of the firewall, does it use a different address block, say part of a /30? mnathani: fwiw i have a /32 for adsl, with a routed /29 at home the internal lan interface on modem is on the /29 ppp makes it easy to have a /32 though but from my understanding, people using hardware firewalls now days often have 3 ip addresses for firewalls, and it bridges one for each firewall, and one for master ip because having a firewall decreases your reliablity, and people who want to spend money on firewalls tend to want to spend money on redundancy there's actually lots of ways of doing things i kind of like routed /32s myself. That's what she said!! where you just route internet ip's onto rfc1918 address space. so your gateway can be rfc1918 even but you're bound to an internal ip err bound to an internet ip for programs that need it so like your webserver maybe on 192.168.6.20, your mail server on 192.168.6.21 your internet ip 4.1.1.1 your mail 4.1.1.2 and you route 4.1.1.1 to 192.168.6.20 and add 4.1.1.1 as a seconday interface on 192.168.6.20 and you set programs to bind to that ip that need it which also means you cna do fancy things like have the ip's bounce between hosts. and have a vpn to access 192.168.6.0/24 the other thing is that you can do /31s more often now too and so you could link using /31 to wan but it tends to kill the odd/even low/high thing