can we start a not-for-profit, community run ISP?
is that something you can do?
@lizardsquid I've been looking into related questions in the context of mesh networks with exit nodes that give the mesh access to the rest of the Internet. I don't know anything about legal/telecom conditions in Australia, but the big telecoms in the US don't like that idea, which means looking more carefully at what kind of business arrangements ISPs make with each other. (If all goes well, I'll soon have published some background on that topic in an upcoming article.)
@jamey @lizardsquid peering at the local IXP and buying transit from 2 (or more) providers will solve those issues.
IXPs are great for connecting to local networks that aren't run by incumbent telcos / jerks (but I repeat myself ;) ).
getting IPv6 is trivial. IPv4 is slightly harder, and can be pretty expensive depending on your region. Looks like AU gives a /22 to new participants, which solves the bootstrap problem.
@phessler @lizardsquid Hooray, that confirms part of my research 😁
I see Hurricane Electric offers IP transit here in Portland and also in Sydney (http://he.net/ip_transit.html), for example… How do you find local transit providers? https://www.peeringdb.com/ is awesome but that's different.
The hard part of launching an ISP seems like connecting customers, anyway. I looked at becoming a DSL reseller, which here gets the Public Utilities Commission involved.
@jamey @lizardsquid HE is a low-cost provider, so just pay attention. And, on IPv6 they don't connect with Cogent so there will be a bit of partition there.
Check out what datacenters are in the city you're in, and they should all have a list of what providers are there.
Peeringdb solves the IXP and peering partner question, which will help you lower your transit costs and get better latency to networks that are close by.
@lizardsquid @jamey and unfortunately, if you're in the US / ARIN region, IPv4 is *very* expensive. ARIN has issued all IPv4 that they have, so you'll have to buy it on the open market. $10-20 per IP, must buy in blocks of 255 (/24 or larger).
You'll need to have at least one /24 v4 block to do anything for real :/
@jamey @lizardsquid twitter, amazon, tumblr, etc, lots of sites are still v4 only.
nat64/dns64 at the network edge will take care of that for most things. skype, corporate vpns, and a very small number of other things still break :(.
thankfully, all of the above can be done with open source, so licensing fees won't be a pain :)
@phessler Okay, those are really helpful observations!
Do those apps also break on a dual-stack network where IPv4 addresses are only from non-routable ranges and then NAT'd the usual way at the network edge? I'd have thought Skype et al had to figure out NAT traversal anyway, so is it just that their client apps can't deal with a v6-only network?
Oh, I guess DNS64 would fight with DNSSEC too, hmm.
Thanks so much for indulging my questions (and I hope @lizardsquid is still interested 😅)!
@jamey @lizardsquid a dualstack network with private v4 would be fine, just a bit more work on the provider side. heck, your home is setup like that.
nat64 is different enough, where these apps freak out somehow :(. I haven't fully looked to understand why.
@jamey @lizardsquid dnssec.....is kinda useless :( it isn't verified on the client, only on the resolver. do the check, and as long as the resolver is careful about when it adds the v6 addr, it's fine.
there is a well known nat64 ipv6 prefix, and if you put the destination there, then a paranoid client ought to be able to verify it.
@lizardsquid @phessler I was wondering that too, but forgot to ask. There are still a lot of servers out there that are v4-only, so you need some kind of interop plan, but in principle I like the idea of a v6-only network with a NAT64/DNS64 endpoint somewhere. (https://en.wikipedia.org/wiki/NAT64)