Posted: 2026-04-16
Not a whole lot of activity on the Slow Servers front lately, though one major feature that many of you could be interested in.
IPv4 is (almost?) available (if you ask nicely I can probably figure it out.) However, I'm not in a rush to spread the message far and wide, because there's also some major upgrades involving downtime on the horizon. Keep that in mind if you sign up!
IPv4 will cost a bit extra, so also keep that in mind.
Currently, the infrastructure is not redundant. If Freya, our router, fails, everything goes down.
The plan is a redundant infrastructure with Thor able to take Freya's place due to software upgrade or downtime for some other reason.
Furthermore, I would like to offer something like "floating IP" -- relayd as a service. This would allow you to have two nodes on two different hosts, exposing a single IP (er, IPs) to the world. Freya and Thor would handle this, and allow for a fully redundant infrastructure. Confined to the limitations of a single datacenter, of course, and a single power drop (my design, not Neutron's!)
If Freya fails, Thor will take over, just in the main router model. And if a host fails, or your VPS fails for some reason, traffic would be routed to the other VPS on another host instead.
But anyway, upgrading the infrastructure to this point will require some downtime. Hopefully minimal, but it could be a couple of hours. Obviously a long time. I plan on scheduling that when I have a date for it.
Once complete, I'll feel better about the state of things and that no single piece of equipment is a single point of failure. It'll also make upgrading routers and hosts much easier.
Deploying IPv4 is another consideration. The current networking model is fully routed and IPv6 only. So each host is responsible for a /56, and each VPS has a /64 (which on new VPSs, is also routed.) This makes the security model pretty solid and straightforward.
This is a little bit tricky with IPv4, though. The most ideal setup might also be fully routed, but this would waste a lot of IPs. Perhaps a /31 per VPS and a /28 or /29 per host.
What I've been leaning towards, unless I figure something else out, is tunneling IPv4 over IPv6, straight to Freya.
There are two methods that seem to work. GIF tunnels and Wireguard tunnels. GIF only seems to work on OpenBSD to OpenBSD. It works pretty well and much lower CPU overhead than Wireguard.
Wireguard works great with both Debian and OpenBSD. However, you burn quite a few extra cycles for moderate traffic loads. In either case, the IP is natively assigned to the VPS and it works just the same other than some MTU reduction.
Speaking of traffic, or traffic amounts, I've realized I may have had an oversight with the bandwidth transfer expectations that I've been setting. It's been 1TB per 1GB of memory, which is simple and is fairly industry standard.
However, there's currently a 100Mbit/sec connection. If you do the math on that, on a 28 day month 100Mbit/sec sustained is 30TB...
Now most people don't come close to using up all of the allocated transfer, but the margins feel a bit tight for me. I am wondering about lowering the transfer amounts, though it might not be a problem at all.
There's been no traffic levels near that 100Mbit/sec threshhold that I've noticed. And upgrading to gigabit is quite possible. However, I don't think Freya or Thor (maybe both together?) are capable of saturating a gigabit link. Perhaps this will improve some with OpenBSD 7.9, but their hardware is definitely on the low end. 2GHz non-speculative Atom, 2 cores, and 2 threads on each.
I could certainly upgrade the routers, though I'm quite fond of the hardware. It's extremely low power and one of the few things on the market immune to speculative execution exploits. Which admittedly are probably a bit overstated, but often worthy of consideration nonetheless. I just like the non-speculative design quite a bit.
For some numbers, going back to Wireguard, I think Freya might hit the ceiling at about 70-80Mbit/sec of Wireguard traffic...
After launching Slow Servers I added SporeStack EU to SporeStack as a new provider offering. It's also whitelabeled hosting.
One nice thing is that I was able to launch it with IPv4 costing extra. I've been able to see just how popular it is to save $1.50/month by not having IPv4. And the results are a little bit discouraging.
Adoption rate of IPv6-only is shy of 4%. Both came up at the same time, and it's easy to select one or the other. I do select IPv6+IPv4 plans by default, but I assume most users explore the dropdown of flavors before picking. I think at one point it wasn't preselected, and there were a number of IPv6-only servers deleted right after launch, to then launch IPv6+IPv4. I think even of my users some of them are confused by it.
This is a little bit discouraging because it leads me to think that I'm missing 96% of the market by being IPv6 only on the Slow Servers side of things. And IPv4 addresses do cost something, but they don't cost a lot. So my whole model of Slow Servers, IPv6-first, is optimized around 4% of possible customer base and pushes away most of the rest of them. Perhaps not the smartest thing from a business perspective.
In practice, the lesser of evils is probably to let people do IPv4 on Slow Servers and fill up some (potentially more) eco-friendly racks that aren't paying into AI infrastructure, rather than have a hardline of no IPv4 because I'd rather not deal with it. I've been using IPv6 since... 2008? maybe earlier. Kinda thought more peoople would've caught up by now, but it's been so slow.