Dynamic Scheduling for Energy Minimization in Delay-Sensitive Stream Mining

Abstract

Abstract-Numerous stream mining applications, such as visual detection, online patient monitoring, and video search and retrieval, are emerging on both mobile and high-performance computing systems. These applications are subject to responsiveness (i.e., delay) constraints for user interactivity and, at the same time, must be optimized for energy efficiency. The increasingly heterogeneous power-versus-performance profile of modern hardware presents new opportunities for energy saving as well as challenges. For example, employing low-performance processing nodes can save energy but may violate delay requirements, whereas employing high-performance processing nodes can deliver a fast response but may unnecessarily waste energy. Existing scheduling algorithms balance energy versus delay assuming constant processing and power requirements throughout the execution of a stream mining task and without exploiting hardware heterogeneity. In this paper, we propose a novel framework for dynamic scheduling for energy minimization (DSE) that leverages this emerging hardware heterogeneity. By optimally determining the processing speeds for hardware executing classifiers, DSE minimizes the average energy consumption while satisfying an average delay constraint. To assess the performance of DSE, we build a face detection application based on the Viola-Jones classifier chain and conduct experimental studies via heterogeneous processor system emulation. The results show that, under the same delay requirement, DSE reduces the average energy consumption by up to 50% in comparison to conventional scheduling that does not exploit hardware heterogeneity. We also demonstrate that DSE is robust against processing node switching overhead and model inaccuracy.

A stream-sensitive distributed approach for configuring cascaded classifier topologies in real-time large-scale stream mining systems

“…Hard deadline is the time limit from where if the processing delay of an SDO (from capturing to coming its final result out of the ensemble) cross, its result will not be useful anymore and it can be considered as a miss (considering a hard delay deadline is common in most of the embedded real-time systems [40][41][42][43]). Hence, a function ( ) must be defined to estimate the proportion of SDOs whose processing delays (from capturing) have not crossed the hard deadline [31,32,44]. Trivially, this function can be given by ( ) = Pr{ ≤ } , which is the probability of the processing delay of a processed SDO (from capturing) that has not exceeded the hard deadline.…”

Media Query Processing for the Internet-of-Things: Coupling of Device Energy Consumption and Cloud Infrastructure Billing

“…4) Billing cost under Pareto, Exponential and Half-Gaussian distribution: We now consider the billing cost for the processing of queries uploaded from n devices via an IoT aggregator. Let us first consider the aggregate query volume distribution modeled via a Pareto distribution with mean E P [Ψ b ] = r tot [with r tot given by (18)], i.e.,…”

“…For instance, when cameras are deployed in a broad square, the data generation volume for some regions will be higher than others, depending on the expected activity of each region [17]. An IoT aggregator can be used to aggregate traffic from each device cluster and upload it to a remote cloud computing service that carries out the search operations for recognition and retrieval purposes [2]- [4], [18].…”

Real‐time face recognition based on pre‐identification and multi‐scale classification

“…After that, the images containing rotating heads were composited using nonlinear manifold learning and finally matched with the synthetic picture. On the other hand, some research studies which integrate spatial and temporal probability models have been on the rise in recent years [4]. Tracking and identification in traditional face recognition algorithms are implemented separately, while in the spatial‐ and temporal‐based method, they are performed at the same time, which is more real‐time but a large amount of time information is needed both in the tracking and identification phases.…”