Modeling the availability of Cassandra

Peer-to-Peer systems have been introduced as an alternative to the traditional client-server scheme. Distributed Hash Tables, a type of structured Peer-to-Peer system, have been designed for massive storage purposes. In this work we model the behavior of a DHT based system, Cassandra, with focus on its fault tolerance capabilities, and more specifically, on its availability when facing two different situations: (1) transient failures, those in which a node goes off-line for a while and returns on-line maintaining its data, and (2) memory-less failures, those in which a node goes off-line and returns with no data. First, we introduce two analytical models (one for each scenario) that provide approximations to the behavior of Cassandra under different configurations, and secondly, in order to validate our models, we complete a set of experiments over a real Cassandra cluster. Experimental results confirm the validity of the proposed models of the availability of Cassandra. We also provide some examples of how these models can be used to optimize the availability configuration of Cassandra-based applications.