Database-Agnostic Transaction Support for Cloud Infrastructures

In this paper, we present and empirically evaluate the performance of database-agnostic transaction (DAT) support for the cloud. Our design and implementation of DAT is scalable, fault-tolerant, and requires only that the data store provide atomic, row-level access. Our approach enables applications to employ a single transactional data store API that can be used with a wide range of cloud data store technologies. We implement DAT in AppScale, an open-source implementation of the Google App Engine cloud platform, and use it to evaluate DAT's performance and the performance of a number of popular key-value stores.