Member
This tracks with my HostHatch experience. Their Zurich location gave me a /64 after I explained I was running LXD and needed routed space. Not instant, not automated, but they understood the request without me having to explain DHCPv6-PD. That counts for something in this market.
Arch on Contabo Nuremberg, /56 working fine with systemd-networkd and a custom [IPv6PrefixDelegation] section. Took me longer to write the network unit than to get the prefix. Their Düsseldorf location is the same setup. No custom kernel, no special sauce.
Has anyone tried forcing a /48 request at any of these providers? RIPE policy allows it but I have never seen a VPS provider hand one out. Curious if Contabo would escalate beyond /56 for a documented multihoming case.
Jar_guru I asked for /48 in my original ticket, they said /56 is the hard limit for VPS tier. VDS and dedicated can get /48 but that is a different price bracket entirely. Their sales were clear about it, no back and forth.
Member
The real problem is not the prefix size, it is the lack of routeability. I have seen providers hand out a /64 then filter ICMPv6 in a way that breaks PMTUD. You have addresses but broken path discovery. Test with tracepath6 before celebrating. Contabo passes, Vultr passes, HostHatch passes. Time4VPS fails. GreenCloudVPS fails intermittently.
Same here on GreenCloudVPS Singapore. /128 only, and they filter ICMPv6 echo but not too big, which is somehow worse than filtering both. Took me a day to figure out why my wireguard tunnel was stalling on large packets. I used https://www.wireguard.com for the setup.
GreenCloudVPS Hong Kong is the same. Single /128, no prefix delegation, support ticket got a "not supported at this time" response. Their Tokyo location might be different but I doubt it. Price is low for a reason.
What about Hetzner? Their cloud product is proper KVM, no excuses about virtualization layer. Yet they only do /64 and you cannot subnet it without breaking SLAAC. I have asked support multiple times, always the same answer: one /64 per server, no delegation.
Member
Hetzner's cloud is KVM but their network design treats each VM as an endpoint, not a router. The /64 is on-link, not delegated. You can use it for containers if you proxy NDP, which is a hack, not architecture. Their dedicated and colo products do proper BGP sessions with your own PI space, but that is a different conversation.
Member
Quiet servers and quiet rooms. Hetzner's Nuremberg hall is loud, their cloud is quiet in the good way. Shame about the v6 policy. I run a /64 with ndppd and it works for my three containers but I know it is wrong.
