Gibbon: An Availability Evaluation Framework for Distributed Databases

Abstract

Driven by new application domains, the database management systems (DBMSs) landscape has significantly evolved from single node DBMS to distributed database management systems (DDBMSs). In parallel, cloud computing became the preferred solution to run distributed applications. Hence, modern DDBMSs are designed to run in the cloud. Yet, in distributed systems the probability of failures is the higher the more entities are involved and by using cloud resources the probability of failures increases even more. Therefore, DDBMSs apply data replication across multiple nodes to provide high availability. Yet, high availability limits consistency or partition tolerance as stated by the CAP theorem. As the decision for two of the three attributes in not binary, the heterogeneous landscape of DDBMSs gets even more complex when it comes to their high availability mechanisms. Hence, the selection of a high available DDBMS to run in the cloud becomes a very challenging task, as supportive evaluation frameworks are not yet available. In order to ease the selection and increase the trust in running DDBMSs in the cloud, we present the Gibbon framework, a novel availability evaluation framework for DDBMSs. Gibbon defines quantifiable availability metrics, a customisable evaluation methodology and a novel evaluation framework architecture. Gibbon is discussed by an availability evaluation of MongoDB, analysing the take over and recovery time.