NoSQL databases are capable of storing and processing big data which is characterized by various properties such as volume, variety and velocity. Such databases are used in a variety of user applications that need large volume of data which is highly available and efficiently accessible. But they do not enforce or require strong data consistency nor do they support transactions. This paper investigates into the transaction processing in consistency-aware applications hosted on MongoDB and Riak which are two representatives of Document and Key-Value NoSQL databases, respectively. It develops new transaction schemes in order to provide NoSQL databases with transactional facilities as well as to analyze the effects of transactions on data consistency and efficiency in user’s applications. The proposed schemes are evaluated using the YCSB + T benchmark which is based on Yahoo! Cloud Services Benchmark (YCSB). Experimental results show that using the proposed schemes, strong consistency can be achieved in MongoDB and Riak without severely affecting their efficiency. We also conduct experiments in order to analyse the level of consistencyof MongoDB and Riak transactional systems.

NoSQL cloud database systems are new types of databases that are built across thousands of cloud nodes and are capable of storing and processing Big Data. NoSQL systems have increasingly been used in large scale applications that need high availability and efficiency but with weaker consistency. Consequently, such systems lack support for standard transactions which provide stronger consistency. This paper proposes a new multi-key transactional model which provides NoSQL systems with standard transaction support and stronger level of data consistency. The strategy is to supplement current NoSQL architecture with an extra layer that manages transactions. The proposed model is configurable where consistency, availability and efficiency can be adjusted based on application requirements. The proposed model is validated through a prototype system using MongoDB. Preliminary experiments show that it ensures stronger consistency and maintains good performance.

No abstract