What is an edge/fog database?

What is it about?

The Edge and Fog Computing paradigms advocate in favor of deploying additional resources closer to application users, rather than performing all computation tasks at a centralized location such as the cloud. Doing so provides advantages in terms of: latency, privacy, network bandwidth, energy consumption, and availability, among others. There is currently no holistic view of how databases fit into this picture. This survey intends to provide a first view into what defines an edge database, what databases exist, and what are the current research axis in the area. It also intends to provide details on the strong and weak points of state-of-the-art systems.

Featured Image

Photo by Jorge Ramirez on Unsplash

Photo by Jorge Ramirez on Unsplash

Why is it important?

We try to bring structure to an area with conflicting definitions. Edge, fog and related terms (e.g., mist and MEC) are often defined differently from publication to publication, with one reference stating they are interchangeable while another defines them as being completely different. As such, this work starts by defining Edge and Fog Computing, and taking those into account to then define edge/fog databases. Systems are divided into application domains, namely: Sensor Networks, Points-of-Presence and Autonomous Vehicles, allowing researchers to better understand the current use case domains of edge database systems, what types of problems they are trying to solve, and which techniques are used to provide an advantage over cloud-based systems. Publications agree on the advantages of edge resources, i.e., lower latency, better privacy, better network bandwidth management and so on. They also agree on its main disadvantage, i.e., that edge nodes often have very limited resources. Concrete values for each of these issues, however, can vary wildly. While the evaluation of one sensor-network database may be using a single core ARM9 microprocessor to represent edge nodes, another may be using an 8-core based server. We provide an analysis of benchmarking and simulation tools, scale and hardware used in systems' evaluation, in order to highlight these discrepancies. On the advantages provided by edge nodes, we analyze latency, energy consumption, and privacy, to identify patterns which, despite evaluation discrepancies, can still be seen across publications, such as the magnitude of latency gain between edge, fog and cloud systems. Specifically for database developers, we also provide a closer look into how common database components, such as the query engine, are modified to adapt to each application domain. Finally, we provide a series of possible future research directions that we hope will help guide future research in the area.