Many cloud based data storage systems, including the menagerie of NoSQL databases that are currently evolving, lack any significant transaction management capability and universally reject the formerly ubiquitous XA standard for classic 2PC ACID transactions. Whilst there are sound reasons for concluding that ACID transactions are not appropriate in such context, this nonetheless leaves a significant shortfall in capabilities compared to mature SQL based storage solutions. As a result, users are forced to roll their own transaction solutions in the business logic layer, an undesirable situation that cries out for a storage vendor or middleware based solution.

In this project you will examine the transaction handling needs of business systems operating on very large quantities of data, compare it with the capabilities provided by the new generation of storage systems used for handling such quantities, and provide solutions to bridge the gap. This may include investigating e.g. the scope for new transaction management standards and APIs that are appropriate or adaptable to more than one storage system; the use of non-ACID transaction models such as compensations; the interoperability issues that arise in an environment that encompasses not only multiple client and server languages (vis. CORBA) but also multiple transports (REST, thrift, ...); the extension of an existing NoSQL solution with additional transaction capabilities, or the provision of equivalent functionality by integration with 3rd party tools e.g. a distributed lock manager.

This project will require a good understanding of Java and optionally one or more additional languages if you wish to extend a storage system built in a non-Java language. Note that languages targeting the JVM are preferred. You should also have a basic understanding of the NoSQL movement and cloud computing concepts such as elasticity. Some knowledge of relevant concepts such as the CAP theorem and BASE model are preferable but not essential. The work will be open source.