Seminar: Simulation Framework for Replicated Data Types Evaluation

Darlington Nwambo
M.Sc. Candidate
Supervisor: Dr. Adrian Fiech

Simulation Framework for Replicated Data Types Evaluation

Department of Computer Science
Friday, December 11, 2015, 2:30 p.m., Room EN 2022



Distributed systems maintain copies of geo-replicated data sets on multiple machines connected via computer networks. Changes made to a data item in a replica residing on one machine is not instantaneously propagated to other replicas (however, the changes should be synchronized eventually). In a per-to-per network, devices communicate with each other (and exchange data) without going through a central server. In such scenario, application developers (e.g. for mobile devices) are burdened by the responsibility to code the synchronization aspect of shared data (e.g. in a calendar application). Replicated Data Types (RDTs) provide encapsulated resolution policies on how to resolve conflicts when replicas on a network/cluster exchange updates. Applications that use RDTs interact with such data types as if only one replica of the data existed (on the particular device running the application). The synchronization of the data is provided by the RDT, happens in the background and is obscured from the application developer. In this project, we implement a Simulation Framework (SF) made up of networks, clusters and devices that will permit researchers to evaluate the role of RDTs in synchronizing data and resolving conflicts. Within a simulation, devices will be joining and leaving networks and device owners will invoke operations provided by the installed applications (which use RDTs). The simulation administrator has the ability to automatically execute scripts that mimic realistic real world scenarios over a specified timeframe.