Trade-offs in causality and concurrency tracking for extreme scale

Participant: Valter Balegas

Home Institution:
CITI - Faculdade de Ciências e Tecnologia

Home Country: Portugal

Host: Prof. Marc Shapiro

Host Institution: LIP6/UPMC & INRIA
Host Country: France

Start Date:

End Date: 2012-10-11

Conflict free replicated data types (CRDT) have been proposed as a technique for allowing a set of replicas to be modified concurrently without coordination, and still guaranteeing the convergence of replicas. These data types have been used to provide data-replication in large scale systems. New designs of these data types provide transactions with a consistency level weaker than snapshot isolation. In the last few years, a large number of research projects have been pushing the limits of transaction memory (TM) in a number of directions. One interesting direction is the application of transactional memory systems in distributed settings, where a set of objects is shared by nodes (typically) in a cluster. In this setting, the latency of communication becomes a major hurdle for efficient transactional memory systems. We envision the possibility of relying on CRDTs, as a form of transactional boosting mechanism for distributed TM systems. In this mission, we intend to make the first steps in the study of the viability of such approach, and if time allows, start the initial design of such a system (or study the integration of these techniques in an existing solution).

Report: here