Edge Clustering Research Articles

DeepZoning: Re-accelerate CNN Inference with Zoning Graph for Heterogeneous Edge Cluster

Parallelizing CNN inference on heterogeneous edge clusters with data parallelism has gained popularity as a way to meet real-time requirements without sacrificing model accuracy. However, existing algorithms struggle to find optimal parallel granularity for complex CNNS, the structure of which is a directed acyclic graph (DAG) rather than a chain, and the parallel dimension is inflexible. To distribute the workload of modern CNNs on heterogeneous devices is also proven as NP-hard problem. In this paper, we introduce DeepZoning , a versatile and cooperative inference framework that combines both model and data parallelism to accelerate CNN inference. DeepZoning employs two algorithms at different levels: (1) a low-level Adaptive Workload Partition algorithm that uses linear programming and takes spatial and channel dimensions into optimization during the search for feature map distribution on heterogeneous devices, and (2) a high-level Model Partition algorithm that finds the optimal model granularity and organizes complex CNNs into sequential zones to balance communication and computation during execution. Our experimental evaluations show that DeepZoning is effective, achieving up to a 3.02 × speed improvement on our experimental prototype compared to state-of-the-art algorithms.

On Optimizing Resources for Real‐Time End‐to‐End Machine Learning in Heterogeneous Edges

ABSTRACTDeploying end‐to‐end ML applications on edge resources becomes a viable solution to achieve performance and data regulations. With the microservice architecture, these applications can scale dynamically, improving service availability under dynamic workloads. However, orchestrating multiple end‐to‐end ML applications within heterogeneous edge environments must deal with numerous challenges while sharing computing resources. Prevalent orchestration tools/frameworks supporting edge ML serving are inefficient in provisioning methods due to constrained resources, diverse resource demands and utilization patterns. In this work, we present a provisioning method to optimize resource utilization for end‐to‐end ML applications on a heterogeneous edge. By profiling all microservices within the application, we estimate scales and allocate them on desired hardware platforms with sufficient resources when considering their runtime utilization patterns. We also provide several practical analyses on runtime monitoring metrics to detect and mitigate resource contentions, guaranteeing performance. The experiments with three real‐world ML applications demonstrate the practicality of our method on a heterogeneous edge cluster of Raspberry Pis and Jetson Developer Kits.

The Possible Aromatic Conjugation via the Different Edges of (Car)Borane Clusters: Can the Relationship Between 3D and 2D Aromatic Systems Be Reconciled?

The possible aromatic conjugation between 3D and 2D aromatic units is in the focus since the synthesis of benzocarborane. It has been showed that in the 3D aromatic icosahedral 1,2-dicarba-closo-dodecaborane systems fused with 2D aromatic rings a global 3D/2D aromaticity does not exist. Despite this fact during the last years several studies proposed interactions between 2D and 3D moieties. Herein, while tuning the size and the effective charge of the (car)borane systems, we demonstrate that global aromatic character can be excluded in any investigated cases, and the detectable conjugative properties can be explained the effect of the well-known negative hyperconjugation.

Reinforcement Learning for Real-Time Federated Learning for Resource-Constrained Edge Cluster

Reinforcement Learning for Real-Time Federated Learning for Resource-Constrained Edge Cluster

Application of Proximal Policy Optimization for Resource Orchestration in Serverless Edge Computing

Serverless computing is a new cloud computing model suitable for providing services in both large cloud and edge clusters. In edge clusters, the autoscaling functions play a key role on serverless platforms as the dynamic scaling of function instances can lead to reduced latency and efficient resource usage, both typical requirements of edge-hosted services. However, a badly configured scaling function can introduce unexpected latency due to so-called “cold start” events or service request losses. In this work, we focus on the optimization of resource-based autoscaling on OpenFaaS, the most-adopted open-source Kubernetes-based serverless platform, leveraging real-world serverless traffic traces. We resort to the reinforcement learning algorithm named Proximal Policy Optimization to dynamically configure the value of the Kubernetes Horizontal Pod Autoscaler, trained on real traffic. This was accomplished via a state space model able to take into account resource consumption, performance values, and time of day. In addition, the reward function definition promotes Service-Level Agreement (SLA) compliance. We evaluate the proposed agent, comparing its performance in terms of average latency, CPU usage, memory usage, and loss percentage with respect to the baseline system. The experimental results show the benefits provided by the proposed agent, obtaining a service time within the SLA while limiting resource consumption and service loss.

Latency-Aware Container Scheduling in Edge Cluster Upgrades: A Deep Reinforcement Learning Approach

Latency-Aware Container Scheduling in Edge Cluster Upgrades: A Deep Reinforcement Learning Approach

Delocalization-ratio analysis of 3-center bonding in position-space for closo-boranes and related systems: Approaching the styx picture and beyond.

Closo-boron hydrides BnHn 2- (n = 5-12) are a conceptually well understood class of compounds. For these and a few related prototype compounds, both the local and the global picture of 3-center bonding are extracted from position-space quantities based on the electron density and the pair density. For this purpose, three-center delocalization indices between quantum theory of atoms in molecules (QTAIM) atoms in position space are used to develop a consistent set of local bond and triangle, and global cluster delocalization ratios (DRs), which are quantitatively compared with conceptual Γ values derived from the styx code for each cluster. Combination of the cluster DRs with associated effective numbers of skeletal electron sharing (SES) for selected cluster surface edges, triangles, or the whole cluster yields effective styx type values describing the trend and even the size of the conceptual styx codes for closo-boranes BnHn 2- and related systems with increasing cluster size n reasonably well. For nonuniform cluster topologies, the different vertex degrees are shown to cause systematic 3-center wise bond delocalization effects for the associated edges and triangles of different average vertex degrees. Extension of DR analysis beyond the styx type triangular cluster-surface bonding corresponds to a triangulation of multicentric bonding. The cluster-wise results keep indicating consistency with the mixed 2- and 3-center bonding approach. The successfully established chemical meaning of the local edge, triangle, and global cluster DRs and their associated SES values constitutes the basis for systematic investigations of mixed 2- and 3-center bonding scenarios in particular in intermetallic and related (endohedral) cluster compounds in the future.

Integrating Entropy Weight and MaxEnt Models for Ecotourism Suitability Assessment in Northeast China Tiger and Leopard National Park

The development of ecotourism in protected areas faces the challenge of balancing conservation and ecotourism. Ecotourism suitability assessments are essential tools for managing tourism in these areas. However, current assessments often overlook biological factors, leading to adverse effects on wildlife. This study uses the Northeast China Tiger and Leopard National Park as a case study to establish a comprehensive assessment system that integrates ecotourism suitability with tiger and leopard habitat suitability, thereby linking ecotourism with wildlife conservation. The primary research methods include ecotourism suitability analysis based on the entropy weight method and habitat suitability analysis using the MaxEnt model. Based on the zoning results of ecotourism and habitat suitability, a comprehensive ecotourism suitability zoning map was produced. This map indicates that areas of very high suitability account for 45.62% of the total area, covering approximately 6152.563 km2, and are primarily located on the edges of village clusters. These areas can be prioritized for developing tourism infrastructure. The comprehensive ecotourism assessment system can balance the development of ecotourism with wildlife conservation, contributing significantly to the coordinated development of economic, social, and environmental objectives.

Foreign Body Reaction to Ion-Beam-Treated Polyurethane Implant.

All artificial materials used for implantation into an organism cause a foreign body reaction. This is an obstacle for a number of medical technologies. In this work, we investigated the effect of high-energy ion bombardment on polyurethane for medical purposes and the reaction of body tissues to its insertion into the mouse organism. An analysis of the cellular response and shell thickness near the implant showed a decrease in the foreign body reaction for implants treated with high-energy ions compared to untreated implants. The decrease in the reaction is associated with the activation of the polyurethane surface due to the formation on the surface layer of condensed aromatic clusters with unbonded valences on the carbon atoms at the edges of such clusters and the covalent attachment of the organism's own proteins to the activated surface of the implant. Thus, immune cells do not identify the implant surface coated with its own proteins as a foreign body. The deactivation of free valences at the edges of aromatic structures due to the storage of the treated implant before surgery reduces surface activity and partially restores the foreign body response. For the greatest effect in eliminating a foreign body reaction, it is recommended to perform the operation immediately after treating the implant with high-energy ions, with minimal contact of the treated surface with any materials.

Collaborative learning-based inter-dependent task dispatching and co-location in an integrated edge computing system

Recently, several edge deployment types, such as on-premise edge clusters, Unmanned Aerial Vehicles (UAV)-attached edge devices, telecommunication base stations installed with edge clusters, etc., are being deployed to enable faster response time for latency-sensitive tasks. One fundamental problem is where and how to offload and schedule multi-dependent tasks so as to minimize their collective execution time and to achieve high resource utilization. Existing approaches randomly dispatch tasks naively to available edge nodes without considering the resource demands of tasks, inter-dependencies of tasks and edge resource availability. These approaches can result in the longer waiting time for tasks due to insufficient resource availability or dependency support, as well as provider lock-in. Therefore, we present EdgeColla, which is based on the integration of edge resources running across multi-edge deployments. EdgeColla leverages learning techniques to intelligently dispatch multi-dependent tasks, and a variant bin-packing optimization method to co-locate these tasks firmly on available nodes to optimally utilize them. Extensive experiments on real-world datasets from Alibaba on task dependencies show that our approach can achieve optimal performance than the baseline schemes.

Genetic network analysis uncovers spatial variation in diversity and connectivity of a species presenting a continuous distribution

The conservation of genetic diversity and connectivity is essential for the long-term persistence and adaptive ability of a species. Recent calls have been made for the inclusion of genetic diversity and differentiation measures in the assessment, management, and conservation of species. However, the literature often lacks direction on how to do so for species with continuous distributions or no distinct breaks in genetic connectivity. There are many considerations to overcome when investigating genetic diversity and connectivity of such species. We combine multiple genetic network methodologies with more traditional population genetic analyses within a single framework to address the challenges of investigating population structure and quantifying variation in genetic diversity and connectivity of wide-ranging species with continuous distributions. We demonstrate the efficacy and applicability of our framework through a study on woodland caribou (Rangifer tarandus) occupying the boreal forest of Canada; a species of significant conservation concern. The dataset consisted of 4911 unique individuals genotyped at 9 microsatellite loci, which were subsequently partitioned into 103 spatial nodes to create a population-based genetic network. The Walktrap community detection algorithm was used to detect hierarchical population genetic structure within the study area and node-based network metrics such as mean inverse edge weight and clustering coefficient were used to quantify the variation in genetic connectivity across the range. Lastly, genetic diversity was assessed by calculating allelic richness and heterozygosity of the nodes making up the network. The community detection analysis identified two communities at the coarsest scale to nine communities at the optimal partition. A strong pattern of Isolation by Distance (IBD) was found across the range at multiple scales. Furthermore, signs of genetic erosion along the study area’s southern boundaries were depicted by nodes presenting low genetic diversity and low centrality values. These results are important to the species status assessments in providing previously unavailable information on connectivity and diversity within and beyond the current local population units used in management. Our approach to quantify the patterns and extent of connectivity across the boreal range is comprehensive and could easily be adapted to other species. The results are robust and provide a solid foundation for the continued monitoring and recovery of the species.

Effects of quenching rate on the plasma gas-phase synthesis of graphene: Insight from the experiments and ReaxFF studies

This investigation analysed the effects of quenching rate on the plasma gas-phase synthesis of graphene. The quenching rate was modulated by the flowrate/type of radial gas whish was injected in the plasma downstream. The experimental results showed that the plasma pyrolysis of acetylene produced spherical carbon nanoparticles. The content graphene flakes in the products increased with the quenching effect, and the layer number of graphene decreased accordingly. Further characterisation confirmed that a high quenching rate increased the crystallinity of the product, reduced the amorphous carbon content, and resulted in better oxidation resistance. The graphene formation pathway was carried out by the molecular dynamics simulations (ReaxFF), focusing on the effect of the quenching rate on graphene growth. As the quenching rate increased, the decrease in the time for surface reactions lowered the generation of five- and seven-membered rings, reducing the possibility of edge growth bending. Besides, an increased quenching rate rapidly lowered the growth temperature and thus retarded the C–H bond breakage at the edges of the carbon clusters. The C–H bonds terminated C–C bond formation at the edges of the carbon clusters, preventing edge growth bending of the carbon clusters, and thus contributing to the growth of lamellar structure. These results suggested that increasing the quenching rate is an effective method for regulating the plasma gas-phase synthesis of graphene.

Probing the Structural and Electronic Properties of the Anionic and Neutral Tellurium-Doped Boron Clusters TeBnq (n = 3-16, q = 0, -1).

In this study, we employ density functional theory along with the artificial bee colony algorithm for cluster global optimization to explore the low-lying structures of TeBnq (n = 3-16, q = 0, -1). The primary focus is on reporting the structural properties of these clusters. The results reveal a consistent doping pattern of the tellurium atom onto the in-plane edges of planar or quasi-planar boron clusters in the most energetically stable isomers. Additionally, we simulate the photoelectron spectra of the cluster anions. Through relative stability analysis, we identify three clusters with magic numbers -TeB7-, TeB10, and TeB12. The aromaticity of these clusters is elucidated using adaptive natural density partitioning (AdNDP) and magnetic properties analysis. Notably, TeB7- exhibits a perfect σ-π doubly aromatic structure, while TeB12 demonstrates strong island aromaticity. These findings significantly contribute to our understanding of the structural and electronic properties of these clusters.

Hierarchical sharding blockchain storage solution for edge computing

The fusion of edge computing architecture and blockchain technology faces challenges such as heterogeneous edge device performance limitations and the inefficiency and high storage redundancy of blockchain systems, resulting in suboptimal fusion results. To address these issues, this paper proposes a layered and sharded blockchain storage model tailored for edge computing. The model adopts a master–slave structure for layered blockchain deployment, which means deploying the main chain in the cloud center to integrate the state information of multiple shard subchains, achieving consistency in the state of the master and slave chains, with each shard subchain responsible for storing data. Meanwhile, considering the performance variations and distribution characteristics of edge devices, this paper designs subchains in the edge chain layer by clustering edge devices with similar performance and proximity into the same type of shard using the concept of community discovery, ensuring balanced storage performance among nodes within the shard. On this basis, this paper also proposes a weight-based consistent hashing algorithm that is capable of distributing data to heterogeneous edge devices according to their weights. Simulation experiments demonstrate that the proposed solution offers advantages in terms of high throughput and low storage redundancy.

Social network analysis of foreign research in the field of automation of cargo container transportation

The relevance of the topic is caused by the following. Improving automation and digitalization of technological processes of transshipment complexes is one of the priorities for development of maritime and river transport in the Russian Federation in the coming years. The importance of scientific research aimed at finding solutions to this problem, undoubtedly, increases already accumulated knowledge and the rate of emergence of new ones in this area. As a result, there is a need to update and systematize newly obtained theoretical and practical developments in review studies, which undoubtedly form the foundation of all academic research activities, as they are aimed at identifying current problems and links of future scientific developments with existing ones. The work analyzes the scientific area research in the field of automation of technological processes of container terminals regarding the dominant thematic areas of foreign scientific schools over the past ten years. The source of information flow for the analysis was secondary open scientific data posted in the GoogleScholar system. The objectives of the study included searching for publications corresponding to the stated topic, clustering the authors of these articles based on the principle of co-authorship, identifying scientists who contributed most to the new scientific developments, determining the scientific interests of researchers by analyzing the texts of their scientific works. The method of edge clustering, as well as structural and positional analyses were used to solve the tasks. As a result, a network graph was generated for the authorship of scientific works, a cross-citation graph of scientific publications, as well as a graph for comparing the research results of scientists with one of the thematic directions of the subject area of research. Graph construction and analysis were performed using Gephi software.

Cometes: Cross-Device Mapping for Energy and Time-Aware Deployment on Edge Infrastructures

As the Internet of Things (IoT) application development falls more on libraries and tools, software developers design applications targeting power-constrained Edge-computing devices. However, not only the Quality-of-Service (QoS) but also energy savings remain major challenges. This article presents Cometes, a complete tool-flow that extends modern SDK tools, assisting developers in selecting resources towards the desired trade-off between execution time and energy consumption, without exceeding the power limitations of the Edge clusters. Cometes runs at application design time, on top of Kubernetes, and exploits cross-device estimation methods as well as traditional multi-objective optimizers to enable efficient resource allocation. The presented solution achieves up to 66% energy savings, up to 50% speed-up and an average improvement of 7%, compared to the placement policy of the lightweight Kubernetes (k3s).

Dynamic edge clustering and task scheduling for edge assisted metaverse system in the field of remote work and collaboration

SummaryThe metaverse, a future digital world for living, working, learning, and interacting, is rapidly gaining significance, particularly in the domain of remote work and collaboration. This emerging digital landscape demands high‐performance, low‐latency, and scalable services to provide an immersive user experience. The current technology used in metaverse systems has some limitations, which emphasize the importance of adopting emerging technologies like Edge Computing (EC). However, as the number of users and data volume increases, it can impact both system performance and scalability of the edge‐assisted metaverse system. Additionally, the uneven distribution of edge servers can cause inconsistencies and result in high latency. To overcome these challenges, this paper proposes a dynamic edge clustering and task scheduling approach for edge‐assisted metaverse systems (DTAM) in the field of remote work and collaboration. The proposed approach addresses the challenges of high user volume and uneven resource distribution by incorporating dynamic clustering and edge server assistance to improve clustering performance. Furthermore, a Prioritized Experience Replay‐based Deep Q‐learning algorithm with state augmentation (PERDQSA) for task scheduling is introduced to improve sample efficiency and performance. The performance of the proposed DTAM is evaluated against existing techniques, and experimental results demonstrate its significant superiority in terms of specific metrics such as bandwidth, task response time, energy efficiency, and latency. The experiments demonstrated that DTAM outperforms Transformation‐based Edge Computing Deep Q‐Learning (TransEC‐DQL) in several key metrics. Specifically, DTAM achieves 28.5% reduction in latency, 13.7% reduction in response time, and 6.4% improvement in bandwidth compared to TransEC‐DQL. These results signify that DTAM can deliver a significantly enhanced user experience in the metaverse, particularly in the context of remote work and collaboration.

Enchancing Lung Disease Classification through K-Means Clustering, Chan-Vese Segmentation, and Canny Edge Detection on X-Ray Segmented Images

The lungs are one of the vital organs in the human body. Not only play a role in the respiratory system, the lungs are also responsible for the human circulatory system. Supporting examinations can also facilitate medical workers in determining the diagnosis. Usually a lung examination is complemented by a chest X-ray examination procedure. This examination aims to see directly and assess the severity of lung conditions. With current technological advances, image analysis can be done easily. Through digital image processing methods, information can be obtained from images that can be used for analysis as a support for diagnoses in the world of health. Image segmentation is a method in which digital images are divided into several segments or subgroups based on the characteristics of the pixels in the image. In this study, clustering with the K-Means method will be carried out on the results of segmentation of x-ray images of lung diseases, namely Covid-19, Tuberculosis, and Pneumonia. The segmentation method that will be implemented is the Chan-Vese Method and the Canny Edge Detection Method. This research shows that the results of the accuracy of applying the K-Means Clustering method to Chan-Vese and Canny Edge-Based Image Segmentation are 80%.

CFL：Data distribution edge clustering algorithm based on deep Federated learning

Federated learning (FL) allows end-devices to train local models on their respective local datasets and collaborate with the server to train a global predictive model, thus achieving the goal of machine learning while protecting the privacy and sensitive data of end-devices. However, simultaneous access to the server by a large number of end devices may result in increased transmission latency and some local models may have malicious behavior, converging in the opposite direction to the global model. As a result, additional communication costs can occur during the federation learning process. Existing research has mainly focused on reducing the number of communication rounds or cleaning dirty local data. In order to decrease the overall amount of local updates, we provide an edge-based model cleaning and device clustering strategy in this study. By computing the cosine similarity between local update parameters and global model parameters over a multi-dimensional space, the approach assesses if local updates are necessary while preventing pointless communications. In addition, end devices are grouped together based on where they are connected to the network and connect to the cloud via mobile edge nodes in clusters, therefore lowering the latency associated with high concurrent server access. Convolutional neural networks and Soft max regression are used, for instance, to perform MNIST handwritten digit recognition, and the effectiveness of the suggested approach to enhancing communication efficiency is confirmed. According to experimental findings, the edge-based model cleaning and device clustering technique decreases the quantity of local updates by 60% and speeds up the model's convergence by 10.3% when compared to the conventional FL approach.

Design of load-aware resource allocation for heterogeneous fog computing systems.

The execution of delay-aware applications can be effectively handled by various computing paradigms, including the fog computing, edge computing, and cloudlets. Cloud computing offers services in a centralized way through a cloud server. On the contrary, the fog computing paradigm offers services in a dispersed manner providing services and computational facilities near the end devices. Due to the distributed provision of resources by the fog paradigm, this architecture is suitable for large-scale implementation of applications. Furthermore, fog computing offers a reduction in delay and network load as compared to cloud architecture. Resource distribution and load balancing are always important tasks in deploying efficient systems. In this research, we have proposed heuristic-based approach that achieves a reduction in network consumption and delays by efficiently utilizing fog resources according to the load generated by the clusters of edge nodes. The proposed algorithm considers the magnitude of data produced at the edge clusters while allocating the fog resources. The results of the evaluations performed on different scales confirm the efficacy of the proposed approach in achieving optimal performance.