DRBD 9 beta enables multiple node high availability
LINBIT, the developer of DRBD, the high-availability distributed block device that has been part of Linux since 2.6.33, has published the first beta of its upcoming version 9. DRBD provides what has been described as a kind of "RAID-1 over the net", which enables secure replication between machines. The disk of a computer, the master, is real-time mirrored to another computer, the slave. This computer can then take over seamlessly when the master fails. To ensure the data is always in sync, master writes are only considered finished when the slave has finished its writing.
The creation of DRBD 9 has involved a major rewrite of the replication solution. Its major innovation is to allow many nodes and many connections between those nodes. This should allow system designers to configure different redundancy mechanisms, for example, to allow for high performance SSDs or optimising integration with cluster managers. Connections between nodes can also be dynamically configured. In previous DRBD releases, a mechanism called device stacking was used to allow multiple nodes to be updated, but this came with some overhead; in DRBD 9 the multiple nodes are addressed using a single activity log, cutting down that overhead. Other improvements in DRBD 9, in part thanks to the major rewrite of the code to enable true multiple node support, include faster random writes, but retain backwards compatibility.
The GPLv2-licensed DRBD 9 Beta (labelled Pre1) is available to download from LINBIT's open source pages. It is unclear when the company intends to submit the new version for merging into Linux; the current beta is labelled as unsuitable for production use.