Interaction-Aware Data Management in the Cloud

Abstract

Cloud computing is a recent trend of technology aimed at providing on-demand Information Technology (IT) services usually priced on a pay-per-use model. One of the main services provided by a cloud computing platform consists of the data management service, or data service for short. This service accepts responsibility for the installation, configuration and maintenance of database systems, as well as for efficient access to stored data. This paper presents a framework, denoted QIDMaC, for management of cloud databases. The proposed framework aims to provide software infrastructure required for the provision of data services in cloud computing environments in an efficiently manner. In this sense, the proposed solution seeks to solve some outstanding problems in the context of cloud databases, such as: query dispatching and scheduling. The proposed approach extends previous work by adding important features such as: support for unpredictable workloads and the use of information about query interactions. Support for the seasonal workloads is related to one of the main properties of cloud computing: fast elasticity. Query interactions can provide significant impacts on database systems's performance. For this reason, QIDMaC uses information about these interactions in order to reduce the execution time of the workloads submitted to the data service and thereby increase the service provider profit. In order to demonstrate the QIDMaC efficiency an experimental evaluation using TPC-H benchmark was performed on PostgreSQL. The results show that the designed solution has the potential to increase the profit of cloud data service providers.