Data Fabrics for Multi-Domain Information Systems

Data exchange in information systems that span multiple policy domains typically rely on network middleware that can abstract the management of underlying heterogeneous communication protocols. This also involves issues in managing interoperability, scalability, and privacy that arise in the movement of data from one domain to another information domain. The Data Fabric is an emerging approach to systematically build and design such middleware systems to support multi-domain exchange at scale. In this paper, we discuss and compare two key data-centric approaches: 1) application layer topic-based messaging and name-based networking in a multi-cloud environment. We implement and deploy these two approaches (using Kafka and NDN) and we compare the performance in terms of object transfer latency and CPU and memory utilization. We find that NDN networking has superior latency performance and lower resource usage. We believe that this advantage derives from the fact that named-based messaging operates at the network level, while topic-based messaging operates at the application level.