Personal tools

A multi-user model for the Anzere storage system

Motivation:

Anzere is a policy-based replication system for personal storage networks. Using Anzere users can specify replication policies such as "Make recent data available at 1 minute latency from my phone" and have them automatically and constantly enforced regardless of changes in the set of personal devices, device failures, data variations, etc. Anzere has so far considered a single-user model and its centralized architecture has been sufficient in such setup. This project investigates how to leverage Anzere to a multi-user model and a distributed implementation that could support the load increase.

Goals:

Get familiar with Anzere and its CLP (constraint logic programming) based implementation. Evaluate possible alternatives for distributing Anzere's replication decisions across different personal storage networks. A potential solution could involve caching CLP decisions and provide reconciliation in case conflicting actions are taken. Design and implement a selected solution, and test its effectiveness in the Anzere overlay consisting of personal devices (mobile phones, PCs, laptops) and cloud infrastructures (Amazon EC2, PlanetLab). Measure the performance improvement compared to the centralized design. Optionally, extend the replication API to include support for multi-user policy specification.