DRBD (LINBIT)

DRBD offers three replication protocols: Protocol A (asynchronous) acknowledges writes as soon as they reach the primary's local disk, Protocol B (semi-synchronous) acknowledges when data reaches the peer's memory buffer, and Protocol C (synchronous) acknowledges only after data is confirmed written to the peer's disk. Protocol C is the default and guarantees zero data loss on failover, making it the standard choice for production clusters.

DRBD 9, released by LINBIT, introduced support for multi-node replication allowing up to 32 nodes in a single DRBD resource. Earlier DRBD 8 versions were limited to exactly two nodes in an active/passive pair. DRBD 9 is particularly useful for three-node high-availability clusters where a tie-breaker node provides quorum in combination with Corosync and Pacemaker.

DRBD core is released under the GNU GPL and is included in the mainline Linux kernel, so it is free to use without licensing restrictions. However, LINBIT sells commercial support subscriptions that include verified enterprise packages, access to DRBD Reactor, and technical support with SLAs. Many production environments running DRBD on distributions like RHEL or SLES rely on LINBIT's commercially supported packages.

DRBD performs synchronous or semi-synchronous replication between servers at the block device level, which is ideal for active/passive clusters where one server directly accesses a replicated device. Ceph rbd-mirror asynchronously replicates RBD images between separate Ceph clusters and is designed for whole-cluster disaster recovery across data centers. DRBD is simpler for two-node HA scenarios like a PostgreSQL primary/standby pair, while Ceph is suited for distributed storage infrastructure.