Site optimal termination protocols for a distributed database under network partitioning

Partition failure causes a major problem for the availability of a distributed database system. No protocol can consistently terminate all parts of a distributed transaction under all possible partitions, since some sites executing subtransactions may not know that other sites have already committed (or aborted) and, therefore, must wait until the failure is repaired. Under a site optimal termination protocol, the expected number of waiting sites is minimized, hence it maximizes the "availability" of a database system in the presence of partition failures. We introduce a new class of partition failures. We introduce a new class of termination protocols, called size-based termination protocols, and identify site optimal termination protocols within this class.