Data Replication for Distributed Emergency Management Systems

Abstract

This paper explores how existing, potentially autonomous applications for crisis management and emergency response can be extended with replication and where nodes are mobile or working in remote areas where they can be offline or weakly connected for extended periods. Tradeoffs between consistency, availability and partitioning tolerance must be made.

The approach investigated provides replication with strong eventual consistency, using operation-based CRDTs, specifically the Last-Writer-Wins (LWW-Element-Set) CRDT. Nodes subscribing to datasets form hierarchical overlay networks. This approach is implemented and tested as an extension to an existing application used in voluntary search and rescue operations. The implementation is backed by an RDBMS and REST APIs. Results are discussed in terms of the LWW strategy, which relies on timestamps. The discussion includes challenges related to delete semantics as well as ways to leverage causality information to enhance the basic LWW strategy.