Objective: Reduce the maintenance burden of rsync mirrors by reducing the number of machines and aiming for higher reliability.
- Keep the masterrsync host around, use it to feed the rest of the network nodes.
- Run a network per continent (us, asia, eu)
- Autoscale the network based on CPU usage
A PoC of this is on GCP.
We have n1-standard-1 instances serving rsync out of a container.
Per region: 2.5 cents/h => month is 18$/month for NLB 25$/mo for a single n1-standard-1 1TB of bandwidth is about 120$