We started assigning IPs to customers using DHCP and giving them a /32 ip and providing a different gw also as a /32. This works fine for the couple of routers we tested on, but now a new customer with limited skills and far away says its not working even if we can see the address is assigned and using a dhcp client in his cpe it works too.
So my question is - has anyone tried this kind with many vendors? Did you see vendors where it dident work?
In this case its a Halon firewall that seems to be BSD based.
Router A is 1.1.1.1/32 so as the ip you add 1.1.1.1/32 with broadcast of 2.2.2.2
Router B is 2.2.2.2/32 so there you use 2.2.2.2/32 and broadcast of 1.1.1.1
To route 0.0.0.0/0 to router A from router B you add a route to 0.0.0.0/0 with gateway 1.1.1.1
Works with point-to-point interfaces, PPPoE would be an example. If an Ethernet interface is moving the packets I would imagine it requires proxy-arp enabled on the WAN interface of the customer router and the router using the gateway. Seems like an awfully over complicated way of doing things. Are you able to provide us with a better view of your topology. I’m thinking some others would be able to provide you some better guidance. Alternatively, switch to PPPoE which should allow this a lot cleaner without creating a large broadcast domain.
If other products got it to work the only reasons I can think of are proxy-arp or the link was set to point-to-point. Alternatively some non RFC compliant mojo. Either way, you’re making it hard on yourself for no reason. If you’re that hard-up for addresses it might be time to think about carrier grade NAT using the shared address space allocation (/10) for it for some or all of your customers.
I never tried with DHCP, but static /32 addressing on othernet works fine (no proxy ARP required) with RouterOS (/ip address add address=1.1.1.1 network=2.2.2.2 interface=), Linux (ip addr add 1.1.1.1 peer 2.2.2.2 dev ) or even Windows (>=Vista, just use 1.1.1.1 as address, 255.255.255.255 as mask and 2.2.2.2 as default gw). But I can imagine that devices expecting “normal” networks with broadcasts and everything might not like it.
The key being the interface reference which tells the OS to ARP on that interface. Without specifying the interface in the route or as the source interface from a ping command I’d be surprised to see it work.
Correction:
peer LLADDRESS
change the link layer broadcast address or the peer address when
the interface is POINTOPOINT.
^^ ip man page in linux. It would seem sets the interface type to pointtopoint which would explain why it works in that situation.