pluto

A typical cloud deployment of an IMAP service follows the service-statelessness principle, i.e. a load balancer distributes the requests to a mostly stateless backend that stores application state on a network file system like NFS. Within this work we suggest an alternative architecture allowing the backend to maintain its own replica of application state, leading to shorter response times and increased fault tolerance. To ensure convergence among replicas, we use CRDTs to model application state and evaluate the performance of our prototypical implementation against the de facto standard IMAP server Dovecot in multiple experimentation settings.