A standard self-hosted requirement is that backups stay cheap while remaining restorable. Here's what works for us.
Layer 1: daily rsync
Every night — rsync -a --delete to two independent VDSs in different regions (Moscow + Helsinki). If an entire DC goes down, the other takes over. We keep the last 7 copies per week.
Layer 2: S3 Glacier beyond 90 days
Once a week, an incremental archive moves to S3 Glacier Deep Archive. Pricing works out to about $0.00099/GB/mo. For a typical 30 GB VDS that is less than a cent per month.
Recovery
SLA: we restore an agent in 30 minutes from rsync, in 2 hours from Glacier (in case both rsync nodes are lost).