On Wed, Jun 25, 2014 at 10:35:38PM +1200, Alexander Neilson wrote:

Hi All

I am looking to see what others experiences are with transit providers.

Mainly I am wondering about IPv6 Support and MTU Sizes

From talking to my upstream providers it appears that 1500MTU is the limit I have access to, and one of my two upstreams doesn?t provide IPv6 for transit.

This leads my to a few logical questions: * How many transit providers provide MTU above 1500 Bytes? * How many transit providers do not provide IPv6 transit? * How do others handle EDNS? I was doing some looking around our systems a while back and I found a default setting in the Bind9 version we were running that set EDNS to 4096 Bytes. Now all of our transit is limited it seems to 1500 Bytes, so I set the config to limit the announced support down into the usable range by us and saw a reduction in the need for DNS retries.

Do other people just have better handling of fragmentation? Do you find any issues having EDNS announcing support for sizes above your transit? do you just have transit that supports MTU over 1500 Bytes?

I'm not a big fan of TCP/IP DNS .. and so I just keep DNS answers small. So I cannot say anything about that sorry. Internet Transit over 1500 bytes is not really doable. Even if you managed to get over 1500 MTU, where are you going to get it to? A lot of the time MTU path discovery doesn't work nearly as well as preferable, so if you have a server with a large MTU you pretty much have to use MSS clamping to fit traffic for 1500 MTU. You may have better luck if you want a larger tunnel between two locations. But the internet in general isn't likely to shift any time soon, as much as I'd like to see it happen. There's also a question of whether large MTU's are even a good thing. On a 10 megabit DSL connection, 1500 MTU means that packets take over 1 msec, to send, and having a few in the queue means you're going to add jitter. If you want to prioritise VOIP traffic or minimise latency in general, then serialisation delay becomes a concern. It's not so bad now that connections are starting to increase in speed, but I don't really see it being a good idea until you get to gigabit speeds, where the benefits of conserving bandwidth hardly matter. And modern ethernet cards deal well with lots of smaller packets.

And a big issue in my opinion is the lack of IPv6 support from transit providers. I am not sure about other providers but it will be one deciding factor when purchasing transit going forward.

I'd rather uptake of IPv6 was controlled and functional myself. If you take a look around the current IPv6 Network most people haven't even configured reverse DNS. The main usage cases where IPv6 is beneficial is in situations like Skype, where clients on shared networks want to create direct connections to each other. Any Skype still doesn't support IPv6 AFAIK. The benefits of web servers, and email servers having concurrent IPv6 and IPv4 is mostly as a kind of proof-of-concept. I wouldn't say it's necessarily a bad idea to choose providers based on being able to offer IPv6, but using it at the moment doesn't really hold any benefit - and that's probably one of the reasons that uptake is so low in New Zealand. Peering is often worse for things like Akamai, and other CDN's. Facebook had a sustained outage that hit (some) IPv6 users and not IPv4 users.

Happy to receive information off list if people don?t want to name on here.

Regards Alexander

Alexander Neilson Neilson Productions Limited

alexander(a)neilson.net.nz 021 329 681 022 456 2326

_______________________________________________ NZNOG mailing list NZNOG(a)list.waikato.ac.nz http://list.waikato.ac.nz/mailman/listinfo/nznog

On Jun 25, 2014, at 5:35 PM, Alexander Neilson wrote:

* How many transit providers provide MTU above 1500 Bytes?

1500 is pretty much the most prevalent MTU size on the Internet these days, as packets ingress at operator edges and then egress them on the 'far side'. So, an MTU larger than 1500 isn't going to help you on the public Internet. It's mainly useful within internal data centers, and WAN-type connections between same.

* How do others handle EDNS?

Fragmentation of large EDNS0 responses (or anything else) is pretty much automagic, from an operational perspective (except when folks have done silly things like block ICMP type-3/code-4 or ICMPv6 type-2). So, this isn't really a concern when looking at the MTU size your upstream transit provider offers. ---------------------------------------------------------------------- Roland Dobbins // http://www.arbornetworks.com Equo ne credite, Teucri. -- Laocoön