Hybrid causality : towards a model between potential and explicit causality

Details

Supervisors

Van Roy, Peter ; Bravo, Manuel

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] : ingénieur civil en informatique

Abstract

Causal consistency has always been considered as the strongest consistency that achieve good availability and throughput. This consistency allows the user to not see any causal anomalies, which means that updates will always be applied after the one that they could potentially depends on. However, new applications, like social networks, struggle with causal consistency and need a consistency that will feel the same to the user, while increasing the performance of the system. To solve this, explicit causality propose to let the application decide the dependencies it wants to use. In this thesis, we will see the model of hybrid consistency, that can use potential or explicit causality in function of what the application require for the operation. We will also present a new system, HybridCassandra, that implement the hybrid consistency while using Cassandra as local datastore. Finally, we will see how to implement this hybrid causality with a simplified version of Facebook. We ran simulation of our system on Grid5000 and we showed that the hybrid consistency outperforms the causal consistency in our system.