CNS Core:Small: Edge Platform for Enabling Situation Awareness Applications

点击翻译按钮获取中文摘要

The ubiquity of affordable computing devices with rich sensing capabilities (e.g., smart cameras, Fitbit, Amazon echo dot, and so on) continues to create increasingly large streams of data collected from the physical world. Situation awareness applications are emerging that utilize these devices to convert sensed information into actionable knowledge. In many instances, the data gathered by these devices far exceeds their computational, storage, and network capabilities. The low-latency and high-bandwidth requirements of next generation applications also means they will quickly become constrained by limitations of the Cloud's centralized design. Edge (or Fog) computing seeks to address this problem by providing Cloud-like capabilities in geo-distributed micro datacenters across the edge of the network. However, realizing this goal is not as easy as it seems due to two challenges. First, micro datacenters are necessarily limited by physical space, and thus have only a small portion of the Cloud’s hardware capabilities. Second, at the same time, there is a sharp increase in demand to process sensed data via machine learning models to achieve rapid actuation. This proposal addresses both of these challenges by introducing a novel set of machine learning libraries (modeled as “Machine Learning as a Service”) to dynamically support different processing requirements across heterogeneous hardware, coupled with an efficient execution model for enabling a high degree of multitenancy on limited hardware resources. The proposed research is aimed at providing insight on how the incoming wave of situation awareness applications may be supported at the edge. Situations awareness applications often require tradeoffs between speed and accuracy when making decisions. At the same time, these applications are continuously evolving, creating a need to dynamically adjust the behavior of a heterogeneous edge computing infrastructure with respect to this tradeoff. In this respect neither the machine learning model used nor the hardware allocated can be static. Micro datacenters at the edge have limited hardware capacity but must provide a high degree of multitenancy to support multiple client applications simultaneously. Thus the execution framework at the edge has to be agile to ensure the most efficient use of limited hardware resources while maximizing performance and availability to different client applications. This proposal explores the research issues pertaining to platform services at the edge (akin to what Cloud computing has done for throughput-oriented applications) that caters to the needs of situation awareness application developers. Planned activities involve in situ studies via collaboration with the Georgia Tech Police Department, which creates the potential for technology transfer and direct societal benefits beyond technical accomplishments.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.