Abstract
The Internet of Things (IoT) is growing at a rapid pace, with billions of devices expected in the near future generating zettabytes of data. The current design of IoT is dependent on the cloud for data storage, processing, and control decisions, which does not scale well to handle this massive influx of data. Edge computing is viewed as a key solution to handle this, and prescribes executing applications closer to devices rather than in the cloud, consequently reducing latency, minimizing bandwidth, as well as improving privacy by operating on premises. The current model of edge computing relies on server-class machines on other public edge infrastructure nearby IoT devices to execute applications, with all control and configuration handled at the cloud. However, having the control plane away from devices, and in the cloud, reduces scalability and reliability, incurs considerable cost, is impractical for deployments without Internet access, and leads to limited data privacy controls for users.
Our work moves away from a cloud-centric design to a device-centric design for edge computing. We identify untapped compute potential in gateway devices present in IoT deployments and utilize it for edge computing, rather than relying on cloud-controlled edge infrastructure. This shift presents several key challenges: handling interoperability of IoT devices, operating on constrained resources, addressing user privacy, and supporting heterogeneous gateways and dynamic workloads. To address these challenges, we first use a decentralized architecture and a thin middleware to enable multiple gateways to operate together, combining their compute capabilities to offer more than the sum of its parts, supporting a good set of edge IoT applications. Further, we create an IoT ecosystem in which resource-constrained IoT devices can offload tasks to more resource-powerful devices, enabling a host of more compute-intensive realtime edge applications to be supported. We also provide users in shared IoT spaces with better transparency and control over their data, by utilizing our gateway-based edge computing platform to enforce user privacy preferences to filter data from edge applications. Together, these solutions enable edge computing which is cost-effective, scalable, general-purpose, and privacy-aware, without the drawbacks of cloud dependency.
