This paper demonstrates ELENNA, an end-to-end latency-aware energy optimization technique to dynamically partition deep neural networks (DNNs) between an IoT device and an edge server. ELENNA employs ML predictors of (i) per-layer compute latency on device/server, (ii) transmission time as a function of intermediate tensor size and network state, and (iii) on-device energy, to determine the partition point that satisfies an application service-quality level (SQL) while minimizing energy consumption. ELENNA is tested in a latency-critical Edge-AI application in the automotive domain. The results show that ELENNA adaptively partitions under time-varying edge server load and network conditions, reducing on-device power draw while keeping end-to-end latency within SQL in edge AI applications.

Mobile devices' processing power and battery life are being strained by the increasing need for real-time computing applications. These tasks can be offloaded to nearby servers with greater processing capability thanks to mobile edge computing. However, choosing the optimal number of tasks to be offloaded to save energy while optimizing the end-to-end latency becomes challenging. This is because we have to make a compromise between minimising energy consumption and optimizing the end-to-end latency. This paper proposes an energy-efficient Deep Neural Network Dividing Technique (EDNNDiT) for latency optimisation in dynamic mobile edge networks. EDNNDiT utilizes external machine-learning models to predict the latency and energy consumption of each DNN layer. These external machine learning models gather real data metrics from the IoT device, edge server, and network components to predict the end-to-end latency and energy consumption for each layer of the DNN model. Next, we use a straightforward heuristic to identify the candidate dividing points of the DNN model that meet the required Quality Service Level (QSL) of the end-to-end latency. Finally, EDNNDiT selects the candidate dividing point with the minimum energy consumption to save power on the IoT device while satisfying the QSL of the end-to-end latency. We conducted real experiments using the NVIDIA JetRacer Robot AI car and Se-QaM platform with both WiFi and 5G that were deployed in our lab under various load conditions on the edge server device. The experimental results for both sequential and non-sequential DNN architectures demonstrate that our EDNNDiT approach reduces energy consumption on the NVIDIA JetRacer while maintaining an acceptable QSL for end-to-end latency. EDNNDiT dynamically adapts to various network conditions and server requirements, showing a powerful and energy-efficient solution for deploying DNNs on limited-resource devices with enhanced overall performance.

Effective end-to-end service quality management is critical for successfully adopting edge computing. However, existing solutions lack the necessary integration of key characteristics to identify and provide actionable insights for resolving the root causes of service quality issues. To address this gap, this paper introduces a service quality manager (SeQaM) to improve service quality in edge computing. SeQaM includes distributed observability, adaptive data collection, real-time analytics, rapid feedback mechanisms, and the ability to create controlled events and experimental scenarios. These features are of utmost importance for infrastructure and service providers, application developers, and researchers to implement, test, validate, and benchmark solutions focused on service quality. To accomplish this, SeQaM is composed of distributed and central components. The distributed components are responsible for collecting service quality metrics and implementing feedback mechanisms in edge applications, user devices, network devices, and edge servers. The central components aggregate the collected metrics, perform holistic analysis, and plan corrective actions to enhance service quality. Moreover, SeQaM can be seamlessly deployed across diverse environments, including emulated testbeds, laboratory settings, and real-world infrastructures. Finally, the effectiveness of SeQaM is demonstrated through three use cases, highlighting its capability to provide detailed insights on the causes of service quality issues, generate data for model training, and support data-driven decision-making.

Dynamic batching in machine learning (ML) serving systems can significantly improve throughput, yet it also introduces a non-trivial trade-off between inference latency and system energy consumption. This paper presents Energy-Latency Trade-off Optimization (ELTO) for ML Inference with Dynamic Batching. ELTO empirically profiles the latency and energy characteristics of any newly deployed model under varying batch sizes and request rates using NVIDIA Triton Server. It leverages supervised regression to predict per-batch latency and energy based on profiled data. ELTO formulates and solves a cost-based optimization to select the batch size that minimizes a weighted sum of normalized predicted latency and energy for deployed ML models. Experimental evaluations using ML vision models (ResNet18, ResNet50) on the NVIDIA GeForce RTX 3060Ti demonstrate the effectiveness of the proposed ELTO. ELTO is compared to heuristic baselines like fixed small or large batch sizes, ELTO significantly reduces the average scaled operational cost-a balanced measure of both latency and energy. For instance, evaluations show cost reductions of $\mathbf{7 5. 5 \%}$ for ResNet18 and $48.2 \%$ for ResNet50 relative to a no-batching strategy, thereby ensuring a more consistently near-optimal operational balance under diverse load conditions

In the Tactile Internet-based fog computing architecture, the sensor devices represent the basic elements for sensing the surrounding environment. They gather a large amount of data due to their use in various real-world Tactile Internet applications. The huge amount of transmitted data from sensor devices to the fog gateway then to the cloud would lead to high data traffic over the network, increased consumed energy, and increased delay to provide the decision at the Fog gateway. These challenges represent a hurdle in the Tactile Internet-based fog system. This paper suggests a Data Transmission Optimization Scheme (DaTOS) in Tactile Internet-based Fog Computing Applications. The protocol works on two-level devices in the Tactile Internet-based fog computing architecture: sensor devices and fog gateway. The DaTOS implements a Lightweight Redundant Data Removing (LiReDaR) Algorithm at the sensor devices level to lower the gathered data before sending them to the fog gateway. In fog gateway, it executes a Data Set Redundancy Elimination (DaSeRE) approach to discard the repetitive data set resulting from the spatial correlation among the data readings sets of sensor nodes. To evaluate the performance of the DaTOS, it was compared to its counterpart methods in the literature like ATP, PFF and Harb. Simulation results indicate that DaTOS outperforms these methods in terms of transmitted data, energy consumption, and data accuracy.

The need to improve smart health systems to monitor the health situation of patients has grown as a result of the spread of epidemic diseases, the ageing of the population, the increase in the number of patients, and the lack of facilities to treat them. This led to an increased demand for remote healthcare systems using biosensors. These biosensors produce a large volume of sensed data that will be received by the edge of the Internet of Medical Things (IoMT) to be forwarded to the data centers of the cloud for further treatment. An edge-fog computing-enabled lossless electroencephalogram (EEG) data compression with epileptic seizure detection in IoMT networks is proposed in this article. The proposed approach achieves three functionalities. First, it reduces the amount of sent data from the edge to the fog gateway using lossless EEG data compression based on a hybrid approach of$k$-means clustering and Huffman encoding (KCHE) at the edge gateway. Second, it decides the epileptic seizure situation of the patient at the fog gateway based on the epileptic seizure detector-based Naive Bayes (ESDNB) algorithm. Third, it reduces the size of IoMT EEG data delivered to the cloud using the same lossless compression algorithm in the first step. Various measures implemented to show the effectiveness of the suggested approach and the comparison results confirm that the KCHE reduces the amount of EEG data transmitted to the fog and cloud platform and produces a suitable detection of an epileptic seizure. The average of compression power of the proposed KCHE is four times the average of compression power of other methods for all EEG records ($Z, F, N, O$, and$S$). Furthermore, the proposed ESDNB outperforms the other methods in terms of accuracy, where it provides accuracy from 99.53 % up to 99.99 % using the data set of Bonn University.

Abstract The necessity of developing sufficient systems to monitor health conditions has increased due to the aging of the population and the prevalence of chronic diseases, creating a demand for remote health care systems that make use of biosensors. This article proposes an energy‐saving multisensor data sampling and fusion with decision‐making for the monitoring of patient health risk in wireless body sensor networks (WBSNs). The work consists of three steps: energy‐efficient sampling rate adaptation, multisensor data fusion, and decision‐making. The sampling is performed in each biosensor and it adapts its rate based on the local risk and the global risk in which global risk computed at the coordinator, where the data is fused afterward. Finally, decisions are made according to the risk level of the patient. The processing of these functions enables in real‐time the adoption of the biosensor sampling rates based on the dynamic risk level of each biosensor, and a corresponding decision is made whenever an emergency is detected. The performance of the suggested approach is evaluated using actual health datasets, and some of its aspects are put into comparison with an existing approach, such as the data reducing and energy‐consuming rates. The acquired results illustrate a decrease in the volume of gathered data, thus a significant energy saving has been made while preserving data accuracy and integrity. Moreover, presenting a data fusing model at the coordinator level by means of an early warning score system has assessed the health condition of patients and took an appropriate decision when detecting emergencies.

The sensor nodes in IoT are usually supplied by energy using batteries with limited capacity. Therefore, saving energy as much as possible is important for increasing their lifetime and thus allowing their use in real applications. Because radio communication is usually the major reason of energy consumption, among the most commonly adopted energy-saving methods is to reduce data transmission / reception, for example, by compressing data. Taking advantage of the high correlation which usually occurs between successive samples captured by IoT sensor nodes and using the entropy principle in compression, a simple lossless algorithm for compression built on Differential Pulse Code Modulation (DPCM) was suggested in this paper, that is especially useful for IoT sensor nodes that characterized by reduced memory and computational resources. Compared to the state-of-the-art approaches, the findings of our simulation experiments showed that the suggested solution substantially decreased the consumption of energy and enhanced the network lifetime for all sensor nodes.

The wireless sensor devices of the Internet of Things (IoT) networks will represent one of the most providers of the big data on the network because it is implemented in the widespread of real-world applications. The large volume of gathered data from the sensor devices leads to increase the communication overhead and thus decrease the limited lifetime of the sensor devices of IoT. Therefore, it is necessary to clean and reduce the redundant sensed data to minimize the cost of communication and save the energy of sensor devices. In this paper, a Data Reduction and Cleaning Approach (DaReCA) for Energy-saving in Wireless Sensor Networks (WSNs) of IoT is proposed. This approach is based on two-level of data cleaning and reduction: the sensor level and the aggregator level. In the latter, we implement a divide and conquer method to merge the near similar data sets which are received from the sensor devices and reduce the transmitted data sets to the sink. In the former, the sensor node will employ a cleaning algorithm based on the leader cluster algorithm to remove redundant data from the sensed data before sending them to the aggregator. The proposed approach is evaluated and implemented using real sensed data of wireless sensor devices with the OMNeT ++ network simulator. The proposed DaReCA approach can clean and reduce the sensed data and save energy whilst keeping suitable data accuracy.

The devices that interconnected to the Internet of Things (IoT) will continue to grow exponentially, and in addition, the amount of data that they report. Sensor nodes (SNs) that arranged in WSNs will create some of IoT data and transmit their readings to Gateway (GW), which driving the sensor nodes to quick expenditure their energy and storage. The low costs of SNs impose a restriction on their energy and storage. To handle these problems it's prefer to carry out reduction on data at the source nodes to reduce both of utilized storage and consumed energy. A large portion of proposed solutions implement data reduction just at one level of the IoT design (e.g. at gateways). A Two-Tier Data Reduction (TTDR) technique is proposed to work at two tier of the network that are: sensor nodes and the gateway. At the sensor node tier we use a simple and suitable data compression methods for constrained IoT sensor nodes. The techniques exploit the temporal correlation in sensor data and use Delta Encoding followed by Run-Length Encoding (RLE). At the gateway tier we apply the hierarchical clustering for grouping data sets received from sensor nodes dependent on the Minimum Description Length (MDL) principle. If any pairs of received data sets can be compressed by the MDL principle, they will be combined into one cluster. Consequently, the amount of data sets is decreased gradually, and the merging of sets in clusters is stopped if the discovery of any match of sets to compress is impossible. Finally, the TTDR performance is evaluated based on real sensory data and using OMNeT++ simulator. The acquired outcomes illustrate the proficiency of the proposed system in regarding data transmission and energy.

In the Internet of Things (IoTs) future, the Wireless Sensor Networks (WSNs) represent one of the big data contributors due to the wide range of real-life applications that use this type of networks. The data volume increases in unexpected ratio. The dense WSN can lead to an increase in the redundant data in the gathered measures of the sensor node. Therefore, it is essential to apply energy-efficient data aggregation to remove the data redundancy and maintain a suitable rate of accuracy. This paper proposes an Integrated Divide and Conquer with Enhanced K-means technique (IDiCoEK) for energy-saving data aggregation in WSNs. The IDiCoEK aggregates the measures in two levels: the node and cluster head levels. A divide and conquer algorithm is applied at the sensor node to remove the redundant data from the collected measures and then send it to the cluster head. The cluster head applies an enhanced K-means approach for clustering the received data sets from the sensor nodes into groups of near similar sets and then the best representative set will be sent to the base station from each group. The IDiCoEK performance is assessed using OMNeT++ network simulator with real data readings of sensor nodes. Results demonstrate that our IDiCoEK technique can save energy by decreasing the measures sent to the sink whilst conserving a suitable level of data accuracy at the sink node.

Popularity of wireless sensor networks (WSNs) is increasing day a day where hundreds or thousands of applications are explored. In most of such applications, the need of gathering data periodically about the monitored environment beside the limited, generally irreplaceable, power sensor sources make energy conservation and big data gathering reduction two fundamental challenges in such networks. In this paper, we propose an Adaptive Distributed Data Gathering (ADiDaG) technique for saving energy in periodic WSN applications. ADiDaG works into rounds where each round consists of three phases: data gathering, sampling decision, and transmission. These phases respectively use Map reduce, longest common subsequence similarity and grouping approach in order to search data redundancy and adapt sensor sampling rate at each round. The performance of ADiDaG is evaluated based on both simulation and experimentations where the obtained results show significant energy savings and high accurate data gathering compared to existing approaches.