A Performance Study of Geo-Distributed IoT Data Aggregation for Fog Computing
Shigeru Imai, Carlos A. Varela, Stacy Patterson · 2018
We investigate MapReduce-based data aggregation for Internet-of-Things data in a multi-tier, geo-distributed datacenter architecture. Specifically, we consider 1) end-to-end hierarchical data aggregation and 2) query response for aggregated data requests made by geo-distributed clients. We first develop a realistic performance model based on previous empirical studies. We then study application performance for various deployment architectures, ranging from a purely cloud-based approach to a geo-distributed architecture that combines cloud, fog, and edge resources. From simulations created based on U.S. Census data, we characterize the trade-off between end-to-end data aggregation time and query response time. Our experiments show that for data aggregation, a purely-cloud based deployment is 53% faster than a deployment with edge resources; however, for query response, the edge approach is 46% faster due to the edge resource proximity to query clients.