@feld

>increasing latency
For bulk traffic, 30-40ms of increased latency is not an issue. For latency sensitive traffic, I use geographically closer datacenters so the latency overhead is at max 8ms for those. I can use policy based routing to pick which VMs (internal IPs) go where.

>throttled pipe
see image. I got four of these bad boys

>nomadic backend
yep that was the idea. I don't have to pay for IPs, if there's a DHCP problem I don't care about DNS, and all my traffic is proxied and cached where needed to cut down on ISP traffic, while I maintain data sovereignty and the cheaper overall resource cost of rolling your own big fat servers.