Transactions on partially replicated data based on reliable and atomic multicasts

We focus on a partial data replication model in which each object is made fault-tolerant by a process group. We propose a replication control protocol that provides transaction properties based both on reliable and atomic multicast primitives and a two phase locking protocol. Furthermore, we propose two deadlock prevention rules to obtain two variants of the base protocol, one that favors the execution of transaction write operations over read operations and another one that does the contrary, favoring read over write. The latter variant ensures that read-only transactions are never aborted. Both protocols ensure that all transactions are not aborted by deadlock prevention rules