Thomas describes MARS as, "an asynchronous replication system for block storage over long distances. It is the base for HA over long distances (more than 50km) and over network bottlenecks, e.g. high / varying packet loss rates." MARS is short for Multiversion Asynchronous Replicated Storage. The project is described in much more detail via its PDF manual.
MARS was originally posted on the Linux kernel mailing list in 2014 but has since been revised so Thomas sent out fresh patches today on the mailing list. MARS continues to already be used in a production capacity at 1&1 Internet where it's said to be used on more than 1,600 servers with more than six petabytes of storage and more than six million operating hours.
MARS comes in at over twenty-seven thousand lines of new kernel code. The patches published today are just for review and it still looks like it could be some time before the functionality lands in the upstream Linux kernel. More details via the kernel mailing list. The out-of-tree module continues to be developed via this GitHub repository.