Performability Model for Assessing NoSQL DBMS Consistency

Carlos Gomes, Eric Borba, Eduardo Tavares, Meuse Nogueira de O. Junior

Abstract: Over the last decades, NoSQL database management systems (DBMS) have been adopted as a prominent technology for big data applications. NoSQL provides high performance and scalability, but there are trade-offs for meeting these non-functional requirements. A major issue is related to eventual consistency between database replicas to achieve improved performance and availability. Usually, NoSQL DBMSs allow the adoption of distinct consistency levels to mitigate problems related to outdated information, though raising the consistency level implies in loss of performance. Such a trade-off may not be trivial to be assessed, since some consistency levels may waste or overuse system resources without obtaining better performance or availability. This paper presents a technique based on generalized stochastic Petri nets to assess NoSQL DBMS consistency regarding latency and availability. A case study is presented to demonstrate the influence of consistency and replication factors on system performance.