Redpanda's development has focused on creating a fast, safe streaming data platform, choosing the Raft replication protocol over Apache Kafka's ISR for its strong consistency model. The importance of fsync, which ensures data is written to disk before proceeding, is emphasized to prevent data loss even in replicated systems. The post argues against the misconception that replication alone can guarantee data consistency without fsync, demonstrating how the loss of unsynchronized data on a single node can lead to global data loss, as shown in an experiment using Apache Kafka. While replication improves data availability and durability, it cannot fully compensate for the absence of fsync, especially in non-Byzantine fault-tolerant systems like Redpanda or Kafka. The post concludes that while Byzantine fault-tolerant protocols could theoretically address some issues, their complexity and performance limitations make them impractical for current use, underscoring the necessity of fsync for maintaining data integrity.