Supporting Multiple Data Replication Models in Distributed Transactional Memory

Abstract

Distributed transactional memory (DTM) presents itself as a highly expressive and programmer friendly model for concurrency control in distributed programming. Current DTM systems make use of both data distribution and replication as a way of providing scalability and fault tolerance, but both techniques have advantages and drawbacks. As such, each one is suitable for different target applications, and deployment environments. In this paper we address the support of different data replication models in DTM. To that end we propose ReDstm, a modular and non-intrusive framework for DTM, that supports multiple data replication models in a general purpose programming language (Java). We show its application in the implementation of distributed software transactional memories with different replication models, and evaluate the framework via a set of well-known benchmarks, analysing the impact of the different replication models on memory usage and transaction throughput.

Simulation of partial replication in Distributed Transactional Memory

“…Servers run a Debian GNU/Linux v6.0.10 operating system and a Java Virtual Machine v. 1.6.0 26. Evaluation was conducted using the ReDstm framework [9] with the SCORe partial replication multi-version algorithm. JGroups [21] provides the underlying group communication service.…”

“…In most systems (e.g. [8], [9], [12]- [14]) byte-code is rewritten, changing the underlying virtual machine or compiler to interpret the "@Atomic" method annotation that delimits the transactions. Transactional code usually provides methods to start, abort and commit transactions, maintaining isolation, as found in relational databases.…”

“…Most of the existing DTM solutions follow a full replication scheme where all nodes in the system keep a replica of every object. However, recent studies show that the reduction of data stored at each node provides to the partial replication approach additional scalability gains [9].…”

Partial Replication Policies for Dynamic Distributed Transactional Memory in Edge Clouds

“…Distributed Transactional Memory (DTM) [1,7,10,11,15,17,18,19] extends STM to distributed systems, addressing new issues such as data replication and node failures. It provides an additional abstraction level to the programmer, where traditional distributed programming details (e.g.…”

Boosting locality in multi-version partial data replication

“…Leveraging on the caching mechanism to improve the replicated system's performance raises two main problems, namely: (i) it shall not affect the correctness of the algorithm, i.e., shall not affect the originally provided consistency guaranties; and (ii) it shall not have a negative impact in the freshness of the data observed by the transactions. This paper addresses the challenges of designing an efficient caching mechanism supporting multi-version partial replication protocols, and have the following contributions: (1) the design of a caching mechanism (discussed in §3) targeting replication protocols providing strong consistency guarantees-which generalizes previous work by Pimentel et al [15]-and is malleable to the point of being used with multi-version partial replication protocols that use scalar logical timestamps; (2) a concrete implementation of the proposed caching mechanism (discussed in §4) developed on top of the ReDstm [19] distributed transactional memory (DTM) framework and applied to SCORe [13]; and (3) the characterization of which transactional workloads may benefit from the proposed caching system (discussed in §5).…”