What's wrong with provider doing 1x IPv6 per VPS ?

XsltelXsltel Hosting Provider
edited May 2020 in General

Hello,

The only valid reason I read for /64 subnet is SLAAC to be used with radvd setup. with recent Cloud-init a VPS provider can assign any number of IPv6/IPv4 automatically without any DHCP during system init.

Now my reason is very simple I deal with multiple providers, some of those providers can't/won't provide /48 subnet. some of them provide a /112 of a shared /64 subnet and that puts me with different subnetting for every product line/location.

So I decided to go with the IPv4 mindset of assigning 1xIPv6 ignoring all the RFCs (some big DCs are not giving a damn f*** about assigning /64 per server why shall I do that on a VPS) However I'de like to hear industry experts take on on this issue.

Best regards
Mohammed H

Xsltel OU | A One-man show powered by 250 grams of brain
Offering reliable hosting services, Server management since 2011 and free cPanel hosting since 2020

Comments

  • What I'd like to see is one IPv6 address assigned on the external network interface, and one reasonably big IPv6 prefix that's routed via that address (or the link-local address).

    The routed IPv6 prefix will give you the flexibility to decide how to use the addresses, for example for docker, VPN clients etc.

    Having a /64 available to the VPS and directly assigned to the external network interface (i.e. not routed) is stupid and forces you to use proxy NDP if you want to use the addresses for another interface, docker or VPN.

    On openvz 6 it seems you have to register each IPv6 address before using it. I don't understand why you have to do that. The venet network interfaces they use are pointopoint interfaces which means you don't have to use proxy NDP anyway.

  • Completely agree with @lebuser

    I don't think there's anything wrong with treating v6 just how you treat v4, in terms of running off a single v6 address. I only really care about v6 access/usage, not too concerned on the implementation specifics - just that the site/app/whatever is reachable over v6.

    On the point of providers splitting up a /64, I just don't trust those for anything remotely important. For anything 'production' I want my own /64 because that's the IPv6 equivalent of a dedicated v4 address - while a routed subnet is ideal, a single or handful of addresses within a unique /64 also works fine for me in most cases.

    🦍🍌

  • DreamDream OGServices Provider

    Just how we do it on our Swiss location.

    We provide one Ipv6 address on the installation of the VM which is in a shared subnet and also mimics the Ipv4 address which is provided. This is for the majority of customers good enough, but we also give out /64 subnets if a customers asks which is really rare currently.

    The point is, that most customers are not familiar with IPv6 and how to set it up or use it. Also the most providers are currently all over the place with the implementation of those setups.

    For the /48 aspects I think you got enough IPs with a single /64 Subnet. I mean for what you need 1.208.925.819.614.629.174.706.176 addresses on 1vCore and 1GB Ram virtual machine? :p
    There is also the financial aspect, of course its not like v4 pricing dimensions but it's there. And no customer would pay let's say 2$ more just to have a /48.

    That's my view of the thing currently who knows if it may change in the near future.

  • XsltelXsltel Hosting Provider
    edited May 2020

    @lebuser said: What I'd like to see is one IPv6 address assigned on the external network interface, and one reasonably big IPv6 prefix that's routed via that address (or the link-local address).

    Thank you so much for this suggestion I didn't know I could use link-local to route an assigned subnet and be done with it.
    reading these resources explains a lot
    https://www.edge-cloud.net/2013/08/07/ipv6-link-local-addresses-as-default-gateway/
    https://blog.zivaro.com/need-know-link-local-ipv6-addresses

    along with this
    https://blog.apnic.net/2018/08/10/how-to-calculating-ipv6-subnets-outside-the-nibble-boundary/

    I believe I can assign /120 (256 addresses) outside the nibble boundary routed through link-local address fe80::1

    Xsltel OU | A One-man show powered by 250 grams of brain
    Offering reliable hosting services, Server management since 2011 and free cPanel hosting since 2020

  • FranciscoFrancisco Hosting ProviderOG

    We originally did a /64 on their interface due to OpenVZ, but we've been thinking about changing it up a lot.

    With us wanting to roll out BGP sessions we need users to have a static primary IPV6 address which isn't a thing right now.

    Right now we're thinking of basically assigning everyone a single IP from a shared /64 by default and just static route their /64 to their link-local (fe80). Probably move new orders (and anyone that hasn't even assigned a v6 IP otherwise) to a /48 per service.

    Francisco

  • WSSWSS OGRetired

    @Francisco I want some /64s in a 128M with umm.. extra bbq sauce?

    My pronouns are like/subscribe.

  • FranciscoFrancisco Hosting ProviderOG

    @WSS said:
    @Francisco I want some /64s in a 128M with umm.. extra bbq sauce?

    If you have a legacy OVZ 128 then it'd have been upgraded to 512MB by now for free :)

    It also has a /64 you can assign out of.

    Francisco

  • WSSWSS OGRetired

    @Francisco said:

    @WSS said:
    @Francisco I want some /64s in a 128M with umm.. extra bbq sauce?

    If you have a legacy OVZ 128 then it'd have been upgraded to 512MB by now for free :)

    It also has a /64 you can assign out of.

    Francisco

    thx - $0.45/yr?

    My pronouns are like/subscribe.

Sign In or Register to comment.