DRBD-8.x

DRBD-8.3.9

• Support BIOs bigger up to 128Kbyte. (instead of 32Kbyte, helps high end performance)
• Allow to freeze IO in case no data is accessible any more (no disk, no network connection). Up to now DRBD failed IO requests in such a situation. Also solved are the two possible ways of recovering from this situation (Attach a disk or reestablish the network connection) by restarting the aborted IO requests. *
• Improve scope of the fence-peer handler, so that such handlers can be used that actually kill the own machine. -- Recovery of this scenario works without split brain auto resolution. *
• A resync speed controller. This allows to run the resync at the highes possible speed across network links with varying bandwidth. Important for long distance replication esp. in conjunction with drbd-proxy. *
• Suspend activity-log writing if possible (i.e. when disconnected and the whole device marked as out of sync). *

DRBD-8.4.0

• Introducing mechanisms to better deal with temporary network failures for devices in primary-primary mode
• Offer an improved resync mode, that deals gracefully with thin provisioned storage. Currently a checksum based resync already has this property. The improvement will be a special mode, that just does not rewrite blocks containing only zeros.
• New metadata layout, designed for the upcomming hard disk drives with a sector size of 4K
• Support for devices bigger than 16TByte (64 bit achitectures only)
...

DRBD-9

• More than two nodes in one level (i.e. without device stacking)
• Full data logging
• Grouping devices into volume groups that obey write ordering and share a single connection
...

* ... feature sponsored by a customer