Quality-aware aggregation & predictive analytics at the edge

Abstract

We investigate the quality of aggregation and predictive analytics in edge computing environments. Edge analytics require pushing processing and inference to the edge of a network of sensing & actuator nodes, which enables huge amount of contextual data to be processed in real time that would be prohibitively complex and costly to transfer on centralized locations. We propose a quality-aware, timeoptimized edge analytics model that supports communication efficient predictive modeling within the edge network. Our idea rests on the capability of edge nodes to intelligently decide when and which data to deliver and process in light of minimizing the communication overhead and maximizing the quality of analytics results. We provide mathematical modeling, performance and comparative assessment over real datasets showing its benefits in edge computing environments.

Smartphone Architecture for Edge-Centric IoT Analytics

“…This concept was partly used in [23] to develop a novel method for meteorological data prediction (Sliding Window-Based Support Vector Regression (SW-SVR)). The concept of selective data forwarding was used as proposed in [24], to design a scheme for forwarding data to the cloud with the aim of reducing the communication overhead and data forwarding cost.…”

“…Nevertheless, it is a prerequisite for the system to meet the requirements of standard edge analytics solutions and hence data transmission should be checked. To satisfy this requirement, the smartphone IoT architecture should utilize the process of selective data forwarding [24] which as the name suggests-allows for data to be transferred only when it is utterly necessary. Moreover, as mentioned in previous sections, this architecture would apply the default to fitted style to implement its edge analytic IoT solution.…”

Knowledge-centric Analytics Queries Allocation in Edge Computing Environments

“…The objective is to predict the query execution time for workload management and capacity planning. The delivery of edge analytics involves communication efficient predictive modeling within the edge network [13]. Analytics are derived by models dealing with dynamic optimal decisions for data deliver in light of communication efficiency [27], [6].…”

Edge-Centric Efficient Regression Analytics

“…Such approach enforces nodes to adopt the same regression algorithm, which is not required in our approach providing the flexibility of hiring different regression models in EDs; our approach relies on the prediction performance of local models independently of the adopted regression algorithms on EDs. Recently, approaches for pushing analytics to the edge are proposed [12] either reduced to distributed parametric regression [16] (whose limitations are discussed above) or to selective data forwarding [15], [11], [13]. Specifically, [15] deals with time-optimized data forwarding among EDs and EGs in light of maximizing the quality of RA.…”

“…Motivations & Goals: We envisage an edge-centric RA paradigm, where EDs are employed as first-class RA platforms [15], [9]. Our motivation is based on RA materialized at the edge including e.g., physical sensors (sensing contextual information), mobile EDs for participatory sensing, and Edge Gateways (EGs) interacting with EDs and sensors/actuators, as shown in Figure 1.…”