STANDARDwalkthrough
Replication on the Ring
A key hashes to position P on the ring. We replicate it to the next RF-1 distinct physical nodes clockwise.
Why distinct physical nodes? Because two consecutive vnodes might belong to the same server.
“With RF=3, the key lives on 3 different servers.”
Blindly replicating to the next 2 ring positions could place all 3 copies on one machine, and if that machine fails, all replicas are lost. We walk clockwise, skipping vnodes on already-selected physical nodes, until we have RF distinct hosts.
For rack-aware placement, also skip nodes in the same rack. Cassandra and DynamoDB both use this clockwise-walk strategy.
Per-node storage with RF=3: , . On node failure, the successor already holds the data: zero cache misses reach the database.
Trade-off: RF=3 triples storage cost and write amplification to 345K total node writes/sec.