The Semantics of Progress and Concurrency in Distributed Transactional Memory Systems

Transactional memory (TM) is an alternative synchronization model for shared in memory data objects that promise to alleviate difficulties with lock-based synchronization (e.g., lack of compositionality, deadlocks, lock convoying). A transaction is a sequence of operations, performed by a single thread, for reading and writing shared objects. Distributed TM is motivated by the similar difficulties of lock-based synchronization methods employed by existing distributed control flow programming models such as RPCs [6]. Past efforts in distributed TM ( [2], [6]) have focused on a transactional throughput regarding both how fast transactions run and that a node running a transaction will eventually be able to complete it. The improved transactional throughput has been evaluated on how to design cache-coherence protocols to maintain object consistency and to reduce the number of aborts through contention management and transactional scheduling. However, none of the past works define semantics for distributed TM. Some of the semantics is defined as a programming abstraction with high-level atomic guarantees. Other semantics are low-level models including aspects of implementations [1]. We aim to design a low-level TM model for progress and concurrency in distributed TM. In this paper, we propose a progress semantic with increasing concurrency in distributed TM. We define a strong progressive commitment ordering (CO) that addresses both the strong progressiveness and distributed CO properties. The paper makes the following contributions: 1) We present a formal semantics called a strong progressive CO for distributed TM systems. 2) We present a distributed lazy version management model to increase concurrency.