Hi RIPE, We've just obtained our final /22 of IPv4 PA space, and we're looking at doing things a little differently with regards to trying to maximise the efficiency of usage within that block. While 100% utilisation of IP addresses by customers is not difficult in e.g. broadband networks (with a /32 being routed over a point to point link to each user), it is difficult in ethernet networks (with 2^n size subnets, network and broadcast and router addresses). Essentially we're going to try to have each customer device (for this purpose, the RIPE atlas probe is a customer device) consume exactly the number of globally unique IP addresses that it needs. Because for isolation each customer gets their own VLAN, and because "ethernet" traditionally means subnets with 2^n size and network and broadcast addresses and physical router addresses and VRRP addresses needing to be on that subnet, we end up with a subnet of 8 IP addresses used for a customer device that only will have 1 IP address exposed to the outside world: globally unique range: 0 - network 1 - default gateway vrrp 2 - router-a physical 3 - router-b physical 4 - customer device .. .. 7 - broadcast What I'd like to do, in order to only consume 1 globally unique IP address for the whole setup is as follows: private range, e.g. 10.a.b.c/24 0 - network 1 - default gateway vrrp 2 - router-a physical 3 - router-b physical 4 - customer device .. .. 255 - broadcast with a single globally unique IPv4 /32 (let's say) p.q.r.s routed to 10.a.b.4 in the above example. This means that you'd need to configure on your device, something like (assuming linux): network: 10.a.b.0/24 gateway: 10.a.b.1 alias: p.q.r.s/32 with it configured that p.q.r.s is the default source IP address to use for any outbound packets. This is a relatively 'clean' way of doing things, as it doesn't involve proxy-arp or similar trickery. It should be possible to do this for linux based systems. What do you think? Will the atlas probe support it, or would you consider making that a feature? Thanks James Rice Jump Networks Ltd.