Aggregate Queries in NoSQL Cloud Data Stores

This thesis work describes the design and implementation of an aggregate view maintenance mechanism for web applications in the Cloud. Maintaining consistent views and ensuring fault and partition tolerance is generally hard as the CAP theorem postulates that these three properties cannot hold simultaneously. Nevertheless, web application transactions are often characterized by short-lived transactions touching only a few data items. Specific examples are shopping cart updates, user name and password look-ups, and online visitor statistics. The small sized updates allow for an efficient aggregate view maintenance solution based on incremental updates via change tables. The change table mechanism can be readily implemented at a transaction coordinator node which can carry out the entire computation because of the small-sized view change tables. This thesis elaborates on using the two-phase commit protocol and simple equi-join queries to implement a synchronous and asynchronous version of the change table algorithm. In this work we will argue that the heavy workloads introduced by the synchronous approach can be decreased by relaxing the view freshness requirements. Thus, the performance of the proposed mechanism can be controlled by exploiting the whole consistency range from immediate updates to deferred refreshing which can be carried out at arbitrary time intervals.