JBOD

Running a broker on several independent disks instead of one combined volume.

Each disk is a separate entry in log.dirs and Kafka spreads partitions across them. A failed disk takes only its own partitions offline, and replication on other brokers covers the loss.