Visualizing and controlling convergence in Lasp program execution

Details

Supervisors

Van Roy, Peter

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] : ingénieur civil en informatique

Abstract

Lasp is an experimental tool designed to handle distributed variables and based on a principle of data convergence: consistency between replicas of a distributed data structure is achieved by using specially designed data structures, named CRDTs (Conflict-free Replicated Data Types). CRDTs have the property that replicas are consistent when they receive the same updates in any order. This allows an efficient implementation in which updates are simply disseminated between replicas, achieving data convergence without requiring usually heavy consensus algorithm. This innovative approach offers many benefits such as a good scalability and an incredibly powerful compromise from the CAP theorem thanks to that new concept of convergence. This master thesis objectives are to improve Lasp by adding new tools to it and to increase knowledge on Lasp deep functioning and limits by analysing its behaviours and performances regarding multiple parameters. To achieve this, a dynamic tool to measure and modify convergence was developed with a clear API and measurements were run and analysed to effectively put Lasp under test.