On Enhancing Concurrency in Distributed Software Transactional Memory

Distributed software transactional memory (STM) promises to alleviate difficulties with lockbased (distributed) synchronization and object performance bottlenecks in distributed systems. The existing distributed STM model based on globally-consistent contention management policies may abort many transactions that could potentially commit without violating correctness. To reduce unnecessary aborts and increase concurrency, we propose the distributed dependencyaware (DDA) model for distributed STM, which adopts different conflict resolution strategies based on the types of transactions.In the DDA model, read-only transactions never abort by keeping a set of versions for each object. Each transaction only keeps precedence relations based on its local knowledge of precedence relations.The DDA model that, when a transaction reads from or writes to an object based on its local knowledge, the underlying precedence graph remains acyclic. We propose starvation-free multi-version (SF-MV)-permissiveness, which ensures that: 1) read-only transactions never abort; and 2) every transaction eventually commits. The DDA model satisfies SF-MV-permissiveness with high probability. We present a set of algorithms to support the DDA model, prove its correctness and permissiveness, and show that is supports invisible reads and efficiently garbage collects useless object versions. Our experimental results show that the DDA model outperforms existing contention management policies by 30%-40% in average in high contention environments.