Every fast write moves work somewhere else

Storage engines must choose when to confirm a write's success to the client, balancing speed against durability—returning immediately after copying to memory is fastest but risks data loss on crashes, while waiting for remote replication or multiple server copies is slower but survives more failures. Modern systems using object storage face particular tradeoffs, as they must decide whether local write-ahead logs are sufficient or if additional remote copies are required before acknowledging success. The latency and failure tolerance of any write operation depend entirely on which durability guarantees the system enforces before returning to the client.