3 ms·

I’ve been using Starlink for a few months now. It’s great for the cost. One annoying thing is their super non-standard and kind of broken network setup if you

by wyager 6y ago

I’ve been using Starlink for a few months now. It’s great for the cost. One annoying thing is their super non-standard and kind of broken network setup if you want to use your own router. They have two subnets with different IP ranges (cgNAT and local) on the same Ethernet link. You have to do some weird manual routes, interface resets, and dhcp client configuration if you want things to work reliably.

walrus01 6y ago
The issue you're seeing there, using your own router, is that during a network-unavailability outage, the starlink terminal will continue serving as a DHCP server to whatever you plug into it. When the network is up and you plug a router into it that's a dhcp client, it'll get an IP address in cgnat space and periodically renew it. When they're doing network maintenance/beta downtime/terminal firmware updates, your router will fail to get an IP address in the cgnat space, but the terminal itself will give the router a new lease in 192.168.100.x (you can ping the antenna at 192.168.100.1), and your router will be happy because it has successfully received a lease and can keep renewing it). Then when the network unavailability is finished, your router keeps its 192.168.100.whatever address and the terminal doesn't issue it a new cgnat address. This sort of setup is weird and not normal, so once your default-configuration DHCP client WAN port router is stuck, you need to manually tell it to go get a new lease, or set up your dhcp client in such a way that it always rejects IPs and leases in 192.168.100.x. I can see why they do it, because they want the starlink app on peoples' phones to be able to communicate with the grpc endpoint on the terminal during outages. But they'll clearly need to do something different with the terminal's firmware and how it treats BYOD router customers, because people will expect to connect a ordinary DHCP client and have it 'just work', and resume working after an outage has cleared.
- garaetjjte 6y ago
  Seems weird. I think they could just always assign local address and route all traffic from CGNAT address to it? It would mask real address from connected device, but as it is CGNAT it isn't very useful anyway.
- iam-TJ 6y ago
  Using a very short lease time for 192.168.100.0/24 when CGNAT isn't available would be one way SpaceX could deal with that. The DHCP client would then be frequently renewing the lease. When CGNAT is available the terminal's DHCP server releases 192.168.100.0/24 and issues/proxies the CGNAT lease. The terminal would be the default route so packets to 192.168.100.1 would still reach it and responses can be correctly routed back to the client using the Starlink monitoring application.

by wyager 6y ago

walrus01 6y ago

The issue you're seeing there, using your own router, is that during a network-unavailability outage, the starlink terminal will continue serving as a DHCP server to whatever you plug into it. When the network is up and you plug a router into it that's a dhcp client, it'll get an IP address in cgnat space and periodically renew it. When they're doing network maintenance/beta downtime/terminal firmware updates, your router will fail to get an IP address in the cgnat space, but the terminal itself will give the router a new lease in 192.168.100.x (you can ping the antenna at 192.168.100.1), and your router will be happy because it has successfully received a lease and can keep renewing it). Then when the network unavailability is finished, your router keeps its 192.168.100.whatever address and the terminal doesn't issue it a new cgnat address. This sort of setup is weird and not normal, so once your default-configuration DHCP client WAN port router is stuck, you need to manually tell it to go get a new lease, or set up your dhcp client in such a way that it always rejects IPs and leases in 192.168.100.x. I can see why they do it, because they want the starlink app on peoples' phones to be able to communicate with the grpc endpoint on the terminal during outages. But they'll clearly need to do something different with the terminal's firmware and how it treats BYOD router customers, because people will expect to connect a ordinary DHCP client and have it 'just work', and resume working after an outage has cleared.

garaetjjte 6y ago
Seems weird. I think they could just always assign local address and route all traffic from CGNAT address to it? It would mask real address from connected device, but as it is CGNAT it isn't very useful anyway.
iam-TJ 6y ago
Using a very short lease time for 192.168.100.0/24 when CGNAT isn't available would be one way SpaceX could deal with that. The DHCP client would then be frequently renewing the lease. When CGNAT is available the terminal's DHCP server releases 192.168.100.0/24 and issues/proxies the CGNAT lease. The terminal would be the default route so packets to 192.168.100.1 would still reach it and responses can be correctly routed back to the client using the Starlink monitoring application.