Distributed Software Transactional Memories : A

Distributed Transactional Memory (DTM) aims at introducing a novel programming paradigm combining the simplicity of Transactional Memory (TM)[11] with the scalability and failure resiliency achievable by exploiting the resource redundancy of distributed platforms. These features make the DTM model particularly attractive for inherently distributed application domains such as Cloud computing or High Performance Computing (HPC). In the HPC domain, several specialized programming languages (such as X10 or Fortress) already provide programmatic support for the DTM abstraction. In the context of Cloud computing, several recent NoSQL data-grids platforms expose transactional APIs and rely on in-memory storage of data to achieve higher performance and elasticity. The Distributed Systems Group at INESC-ID has started to perform research on DTMs in 2008. In that year, we have presented at the LADIS workshop a position exposing a number of research directions that we aimed at addressing[18]. Over the last 4 year we have published a number of papers[21], [20], [6], [19], [7], [3], [13], [5], [4], [22], [8], [16] based on prototypes that we have built to explore these research directions, in an attempt to better understand the advantages and limitations of the DTM paradigm. The current paper makes an overview of this research, highlighting the main findings and providing directions for future work.