Abstract

The proliferation of Internet-of-Things (IoT) applications requires new systems to collect, store, and analyze time-series data at an enormous scale. We believe that meeting these scaling demands will require a significant amount of data processing to happen on edge devices. This paper presents VergeDB, a database for adaptive and task-aware compression of IoT data that supports complex analytical tasks and machine learning as first-class operations. VergeDB serves as either a lightweight storage engine that compresses the data based on downstream tasks or as an edge-based database that manages both compression and in-situ analytics on raw and compressed data. By optimizing for available computation resources, storage capacity, and network bandwidth, VergeDB will take decisions to maximize throughput, data compression, and downstream task accuracy.