Ensuring the reliability of edge services is increasingly challenged by the complexity and stochastic behavior of host Internet of Things (IoT) devices. Recent efforts have leveraged formal logics, such as Signal Temporal Logic (STL) and its probabilistic variants, for predictive monitoring of QoS degradations and service failures. However, these approaches lack support for the quantification of satisfaction probabilities for composite services and over time intervals, and remain limited by their reliance on parametric QoS distributions. To address these challenges, this paper proposes Joint Probabilistic STL (JPSTL), an extension of STL that introduces novel metrics for estimating joint probabilities across both compositional and temporal dimensions. Specifically, QoS values are modeled using non-parametric distributions, enabling the construction of stochastic signals predicted via Bayesian neural networks and sparse kernel density estimation. These QoS distributions are thereafter aggregated along compositional and temporal dimensions to support predictive monitoring through JPSTL-based formal verification. Extensive evaluations on real-world datasets demonstrate that JPSTL significantly outperforms state-of-theart techniques in monitoring accuracy while providing robust probabilistic guarantees for complex edge-native applications.

With the rapid development of 6 G communication technology and the Internet of Things (IoT), mobile edge computing (MEC) is regarded as an effective paradigm of providing low-delay, high-quality services to mobile users. In the IoT device-edge-cloud network, the optimal deployment of MEC servers is a prerequisite for a better task offloading, while the improved performance of mobile users task offloading also indicates the deployment scheme is optimal. Most of current MEC servers deployment studies focus on reducing delay and deployment costs, but ignore the offloading requirements of mobile users with similar task type and cooperative relationship arriving at the same community. In this paper, we study the MEC servers deployment driven by the task offloading requirements of community mobile users in current period by utilizing the stability of their social cooperative relationships to maximize the service satisfaction of all community mobile users in the future task offloading. First, the cooperative relationship strength between mobile users is measured to form a group of resource requesters based on interaction probability, movement trajectory and credit strength. Then, we implement the optimal search of base stations (BSs) using spatial index, followed by the one-to-many matching theory between BSs and community group resource requesters, to balance the load of BSs and reduce the communication delay between them. Finally, we use TD(lambda) algorithm and task similarity between cooperative users to deploy MEC servers with suitable resources around BSs so that the deployment scheme can significantly improve the future task offloading performance of all community mobile users. Based on the real data set provided by Shanghai Telecom, it is confirmed that the proposed scheme has significant advantages in improving all community mobile users service satisfaction, with an average improvement of 18.49% compared with the baselines.

Serverless edge computing has emerged as a pivotal paradigm for hosting latency-sensitive applications. However, it remains plagued by severe cold-start delays under stringent resource constraints. While Zygote containers effectively mitigate initialization overhead through dependency sharing, existing deterministic pre-warming strategies are ill-equipped to handle extreme spatio-temporal traffic bursts, which often trigger cascading startup failures and excessive memory fragmentation. To address these challenges, this paper proposes the BurstEdge, an uncertainty-aware Zygote planning framework designed for burst-resilient function execution. Specifically, BurstEdge explicitly quantifies workload volatility via a probabilistic demand predictor, which drives a risk-aware Integer Linear Programming (ILP) formulation for optimal cloud planning. At the edge, an agile dual-pool routing mechanism is devised to dynamically absorb localized traffic surges with millisecond-level responsiveness. Extensive evaluations using real-world Zygote functions demonstrate that BurstEdge significantly suppresses cold-start rates while enhancing memory efficiency and cost-effectiveness, consistently outperforming state-of-the-art methods in highly volatile edge environments.

Edge service monitoring is essential for ensuring the robustness and efficiency of service executions, where predictive monitoring enables proactive detection of potential service violations. Current approaches for predictive monitoring, which mostly adopt Signal Temporal Logic (STL) specifications for requirements representation and evaluation, primarily focus on deterministic signals, and thus, may lack probabilistic guarantees for uncertainty interpretation. To address these challenges, this paper proposes Bayesian STL (BSTL), an extension of STL that enables probabilistic reasoning over stochastic signals. Specifically, Bayesian Neural Networks (BNNs) are employed to generate sequences of posterior probability distributions, offering more comprehensive predictive insights compared to traditional point- or interval-based methods with deterministic sequential predictions. Uncertainty interpretation over these distribution predictions is achieved by a novel expected robustness metric that jointly quantifies both the degree and probability of service satisfaction. Thereafter, a BSTL-based predictive monitoring framework is developed, where a service constraint is formally specified by a BSTL formula and interpreted with both qualitative and quantitative semantics. Besides, confidence levels and constraint thresholds ensuring robust satisfaction of a BSTL formula are rigorously estimated. Extensive experiments on publicly available datasets demonstrate that BSTL outperforms baseline techniques in terms of expressiveness, robustness, and applicability.

In the Internet of Things (IoT), domain applications can be achieved by assembling the functionalities of IoT devices, which can be encapsulated as IoT services. A precise and adaptive monitoring at runtime is fundamental to guarantee the health of running IoT service compositions. Current techniques can hardly conduct the monitoring of dependencies prescribed upon IoT services in compositions. Besides, formal methods, like Signal Temporal Logic (STL) and its last extensions, lacks operators to collectively monitor composite IoT services hosted by spatially contiguous IoT devices for their resource usage fluctuation. To mitigate these limitations, this paper proposes a novel Spatial Composite STL (SCSTL), which enhances STL with spatial encoding and composite operators, to achieve a formal and robust monitoring at runtime, for dependencies prescribed among IoT services, and the resource usage fluctuation of spatially distributed IoT devices. Evaluation results show that our SCSTL outperforms the state-of-the-art techniques.

Nowadays IoT devices in Mobile Edge Computing (MEC) networks have been deployed in large-scale quantities to guarantee sensing data collection for anomalous event detection as full as possible even if some devices are in fault. Some techniques, such as clustering and dimensionality reduction, are adopted to eliminate redundant sensing data collection in this large-scale deployment. However, they not only have high computational complexity and easily cause the loss of information on the primary sensing attributes for detection, but also bring certain errors to the detection because of their low sensitivity to data processed. In addition, insufficient collection of primary attribute data samples often results from physical or human factors, and mindless imputation of large-scale data gaps without basis may lead to greater irreparable losses. To address the above challenges, we first complete the selection of optimal primary attribute device collection and aggregation (PADCA) path based on minimum spanning tree, reducing data communication cost for redundant primary attributes collection. Then, we propose an anomalous impact correlation search strategy to quickly locate all MEC servers whose management regions have cascading anomalous event and help determine the transferable source MEC servers. Leveraging this, we use transfer learning to help detect anomalous events in the management regions of the MEC servers with insufficient primary attribute data samples, where a particle swarm optimization based back-propagation (PSO-BP) neural network model is used to infer the fusion weight of each primary attribute. Experimental results show that our method achieves higher detection performance in terms of detection time, energy consumption, accuracy, and receiver operating characteristic (ROC) curve compared to the benchmarks by at least 24%, 34%, 0.5 and 0.05.

Edge computing facilitates low-latency and high-bandwidth services by deploying edge servers close to data sources and end users. However, dynamic edge environments, characterized by fluctuating resources and user mobility, etc., necessitate efficient service migration to maintain consistently low latency. Traditional migration methods incur significant latency due to redundant data transfers, especially in containerized environments where shareable storage layers and redundant files across non-shareable layers in Docker containers exacerbate inefficiencies. Despite advances in layer sharing migration techniques, the challenges remain as follows: (i) shareable layers are limited due to coarse granularity, and (ii) approximately 97

Containerized edge services must often be migrated when users move, resources fluctuate, or access latency degrades. Existing migration systems mainly reuse complete image layers or relocate containers after an orchestration decision has already been made. They therefore miss duplicated files hidden inside non-shareable layers and transfer unnecessary image and writable-layer data. This paper proposes a file-centric collaborative migration framework for containerized edge services. This framework preserves native layer sharing but extends reuse to file granularity through a layer-preserving file repository, SHA-256 based file metadata, and a pipelined migration procedure that pre-transfers static files and checkpoints before the final stop phase. It further eliminates Copy-on-Write (CoW) redundancy by transferring only new files and modified blocks in the writable layer. We implement and evaluate a proof of concept on a VMbased emulated edge testbed under controlled network conditions. Experiments show that our method reduces total migration time, the stop-phase container-layer transfer time that dominates service downtime, and estimated edge-node transmission energy compared with state-of-the-art approaches.

Continuous multi-batch drone delivery operations introduce progressive battery degradation that causes soft failures, where service quality gradually deviates from expected norms without triggering explicit fault alarms. Existing approaches rely on either physics-based models or data-driven detectors, both of which exhibit stage-dependent reliability limitations under sustained operation. Static fusion strategies fail to adapt as the relative reliability of each paradigm shifts across degradation stages. This paper proposes an adaptive knowledgedata fusion framework for Quality of Service (QoS)-aware soft failure detection in Swarm-based Drone-as-a-Service (SDaaS). This framework employs a dual-track architecture that combines physics-informed degradation modeling with telemetry-driven temporal learning. An uncertainty-aware gating mechanism dynamically adjusts fusion weights based on real-time uncertainty and operational context, emphasizing the more reliable paradigm at each degradation stage. Beyond detection, a severity-guided reconfiguration strategy triggers proportionate recovery actions to sustain service continuity. Extensive experiments on real-world battery degradation datasets and urban delivery simulations demonstrate that the proposed framework achieves superior detection reliability compared to single-paradigm and static fusion baselines.

Service-oriented computing (SOC) relies heavily on service composition and service recommendation to build flexible distributed systems and meet complex user needs. Traditional methods in these areas face inherent limitations, including heavy dependence on expert knowledge, poor adaptability to unstructured natural language inputs, data sparsity, and challenges in handling dynamic service ecosystems. With the rapid advancement of Large Language Models (LLMs), their exceptional capabilities in natural language understanding, reasoning, and knowledge integration have opened new avenues for addressing these long-standing issues. This survey provides a comprehensive analysis of LLM-enabled service composition and recommendation. We systematically review three foundational pillars: LLM-driven frameworks (contrasted with traditional paradigms), high-quality benchmark datasets and four core motivations for adopting LLM-driven approaches. We then synthesize key techniques spanning model architectures and generative planning, and validate practical effectiveness across four typical application domains. Finally, we discuss critical challenges including implicit requirement understanding, reliable execution of LLM-generated workflows, cold-start in cross-domain scenarios, and security, privacy, and trust concerns, along with corresponding future research directions. This survey aims to synthesize existing knowledge, identify research gaps, and guide the development of LLM-enabled service computing systems.

Mobile Edge Computing (MEC) and Device-toDevice (D2D) peer offloading are two promising paradigms in the mobile Internet of Things (IoT). In this paper, we study the collaborative task offloading with redundant data and codes in large-scale IoT networks, where computing resource-starved IoT devices can offload their tasks to MEC servers via cellular links or to nearby peer devices (PDs) with idle resources through D2D links for execution. IoT tasks usually consist of a series of dependent and parallel subtasks, and the difficulties in current research are (i) how to eliminate redundancy in data or codes between subtasks, and (ii) how to leverage previous experience to adaptively search a set of collaborative MEC servers and PDs for matching offloading of dependent and parallel subtasks. From this, we propose a redundancy-aware adaptive search offloading (RASO) method based on the deep Q-network (DQN). Specifically, we first design a fine-grained task recombination scheme by judging the consistency of subtask data and codes. After that, we organize the global devices into a spatial index MP-tree to reduce the search solution space, and propose a fast adaptive search method based on the DQN combined with MP-tree, where optimal path-guiding parameters training of inner and outer layers is involved to efficiently help achieve collaborative devices to complete specific tasks with the same type. After finding the collaborative MEC servers and PDs along MP-tree for a certain task, a centralized stable matching algorithm is further developed to give a decision of offloading each of its divided dependent and parallel subtasks to the matched one, thereby optimizing offloading delay and energy consumption. Extensive simulation results show that compared to other counterpart solutions, our proposed method has improved task offloading performance in terms of delay and energy consumption.

Porphyry copper deposits (PCDs) are the most important deposit types for essential metals of copper, molybdenum, and gold. However, they are becoming increasingly harder to discover after decades of large-scale worldwide exploration. How to build a more convincing mineral system model, by accurately identifying all indispensable theoretical components from ore deposit formation processes, to guide the PCDs exploration during the modern era of diminishing returns on fixed exploration budgets, is the most urgent work. Knowledge graph (KG), by constructing a graph database to store the concepts and entities and the complex relationships between them, has robust knowledge discovery (KD) ability to obtain new knowledge and conclusions from existing data. With the continuous publication of massive research and literature, studies on PCDs have entered a big data era. Therefore, in this study, we try to use the KG and the following KD to develop a robust workflow for knowledge graph construction and discovery, which facilitates the identification of some indispensable and previously overlooked porphyry mineral system components from published articles, preparing for linking with spatial data and exploration budgets. We provide not only a detailed roadmap for KG construction and KD of PCDs, but also a typical example of the construction and application of KG in geosciences.

The heterogeneous resources co-optimization in edge networks is essential to enhance the network throughput. Existing load-sensitive (re-)scheduling approaches mostly formulate the heterogeneous resources balancing as a single-objective optimization issue, omitting the balanced usage of heterogeneous resources on a given edge node. Moreover, these approaches are inadequate for the heterogeneous resources adaptive cooptimization, microservice dependency modeling at a more granular level, and multi-step online re-scheduling. Thus, a Dependency-aware Online Microservice re-Scheduling (DOMS) approach is introduced. In particular, we formulate the microservice re-scheduling as a multiple knapsack optimization issue, and solve it through the Double Dueling Deep Q-Network (D3QN) with prioritized experience replay. Our DOMS incorporates a heterogeneous resources adaptive balancing detection algorithm to enable adaptive co-optimization of heterogeneous resources. A fine-grained dependency graph of microservice performance metrics is built, upon which a multi-step scheduling partition algorithm is devised to facilitate multi-step online re-scheduling. Extensive experiments on a public dataset show that DOMS outperforms comparison approaches in terms of latency, energy consumption, balance degree, and throughput.

Collaborative sensing has emerged as a novel sensing paradigm, entailing multi-sensor data sharing and multimodal modeling to collaboratively understand sensing behaviors. However, current solutions, i.e., data-level and decision-level fusion methods, fall short of generality, expert knowledge, and holistic/chronic perspective. In this paper, we propose LLMCoSen to revisit collaborative sensing with Large Language Models (LLMs). Specifically, LLM-CoSen designs a semantic-level fusion approach for inference results for collaborative sensing. Such an approach is characterized by its generality, making it applicable to any heterogeneous devices, and its expert knowledge incorporation, which provides chronic, holistic, and insightful perspectives on the inference results. Regarding inference absence challenges, we propose a personalized model design method to constrain inference time, and a voting-based two-pass prompt engineering strategy for token completion. Regarding inference error challenges, we propose an accuracy restoration strategy for personalized models, and a two-level error estimator coupled with self-correction. Experimental results of human digital system use case on four corresponding benchmark datasets show LLM-CoSen can decrease inference absence by 72.83% and inference errors by 7.65% on average.

Detection of anomalous event at the edge of network has attracted wide attention from both academic and industrial fields recently. During the detection process, several primary sensing attributes are jointly utilized to determine whether an anomalous event occurs or not. However, as the primary attributes of some Internet of Things (IoT) devices are easy missing due to the natural wear and they cannot be timely and accurately accessed, the event detection efficiency is very low. In view of this, our work introduces a digital twin (DT)-assisted detection technology for anomaly identification in a device-edge-cloud architecture. Specifically, for an edge server with missing primary attributes, the probability of anomalous event occurring on it can be calculated by analyzing the primary attribute fusion values of its adjacent edge servers. As a result, it is unnecessary to carry on detection in advance on the edge servers with a low anomaly occurring probability, efficiently reducing the detection cost. For the remaining edge servers with a high probability, the primary attributes with high accuracy are migrated by considering the difference on the historical value variant trend and the fusion effect. Based on this, a decision tree will be built in the integrated DT model for anomalous event detection in advance. Further, the cloud collects other relevant attributes to build a random forest for the final identification and judgment of anomalous events. Experimental results show that our method achieves a higher detection performance in terms of energy consumption, detection time, and accuracy by at least 37.1%, 39.5%, and 1.82% compared to the baselines.

Internet of Things (IoT) systems gained significant attention for monitoring and optimizing processes. To ensure realtime detections with low latency, IoT applications often monitor individual components through a microservice network, deployed close to IoT devices. Existing methods for multivariate time series anomaly detection typically construct one global graph for identifying deviations in predicted or reconstructed attribute features. However, consider an active node that suddenly experiences a sharp drop in connections or established unexpected links; these structural anomalies would be totally overlooked. To address these limitations, this paper proposes Multi-Granularity Graph Anomaly Detection (MGG-AD), a novel approach that captures both attribute and topological dependencies within IoT systems. Specifically, we construct a multi-granularity dependency graph from a global graph and multiple local subgraphs that define geographical correlations among IoT devices. First, at the attribute-level, we detect contextual deviations by reconstructing feature representations and contrasting attributes across local subgraphs. Second, at the topological-level, we identify abnormal structural variations by comparing local subgraphs with the global graph-called contrastive learning. We evaluated MGGAD on two publicly available datasets and against state-of-the-art methods-we found that our solution provides higher detection accuracy and robustness, underlining its suitability for dynamic IoT systems.

With the rapid proliferation of Internet of Things (IoT) devices and mobile applications and the growing demand for low-latency services, edge computing has emerged as a transformative paradigm that brings computation and storage closer to end users. However, the dynamic nature and limited resources of edge networks bring challenges such as load imbalance and high latency while satisfying user requests. Service migration, the dynamic redeployment of service instances across distributed edge nodes, has become a key enabler for solving these challenges and optimizing edge network characteristics. Moreover, the low-latency nature of edge computing requires that service migration strategies must be in real time in order to ensure latency requirements. Thus, this paper presents a systematic survey of real-time service migration in edge networks. Specifically, we first introduce four network architectures and four basic models for real-time service migration. We then summarize four research motivations for real-time service migration and the real-time guarantee introduced during the implementation of migration strategies. To support these motivations, we present key techniques for solving the task of real-time service migration and how these algorithms and models facilitate the real-time performance of migration. We also explore latency-sensitive application scenarios, such as smart cities, smart homes, and smart manufacturing, where real-time service migration plays a critical role in sustaining performance and adaptability under dynamic conditions. Finally, we summarize the key challenges and outline promising future research directions for real-time service migration. This survey aims to provide a structured and in-depth theoretical foundation to guide future research on real-time service migration in edge networks.

The proliferation of Internet of Things (IoT) applications prompts extraordinary demands for the collaboration of large amounts of computational resources provided by IoT devices in edge networks, and these applications are mostly delay-sensitive. Generally, these resources are encapsulated as IoT services. Thereafter, IoT applications can be performed, such that the collaboration of their sub-tasks is achieved through the composition of functionally complementary and geographically contiguous IoT services. The status of computational resources in IoT devices may change continuously along with their occupancy and release by IoT services. Considering the resource-scarceness of IoT devices, when the workload of IoT devices increases due to more services to be processed, certain IoT devices may hardly have enough remaining resources to co-host more instances of certain IoT services prescribed by forthcoming IoT applications with strict constraints. As a result, the delay satisfaction of both on-running and forthcoming IoT applications may be negatively impacted, or even hardly be satisfied any longer. To solve this issue, this paper proposes a rEsource-Efficient service Configuration ($E^{2}$rC) mechanism, which aims to optimize the configuration of computational resources provided by IoT devices with respect to complex requirements prescribed by IoT applications, through service migration techniques. This service migration problem is formulated as markov multi-phases decisions, which is solved through our enhanced Deep Reinforcement Learning (DRL) approach with a two-layer Q-network. Extensive experiments have been conducted upon the dataset of our testbed system. Evaluation results show that our $E^{2}$rC is more efficient than the state-of-art counterparts in satisfying delay constraints of IoT applications, while reducing the energy consumption and improving the resource utilization efficiency of IoT devices.

Anomaly detection has emerged as one of the core research topics to support workflow applications across domains. To differentiate anomalies from normal patterns of workflows, Graph Neural Networks (GNNs) models have been introduced. These models leverage time series data to construct graph structures, in order to explicitly capture task dependencies among industrial Internet of Things (IoT) devices, and thus, to identify deviations from predicted behaviours as anomalies. However, existing forecasting-based anomaly detection methods may not accurately detect certain anomalies, as they rely solely on historical sensory data while seldom considering the valuable information embedded in domain knowledge. To address this limitation, this paper proposes a Knowledge-enhanced graph attention-based Anomaly Detection (KeAD) method. Specifically, a knowledge-enhanced graph structure is constructed by incorporating domain-specific knowledge to represent spatio-temporal dependencies between IoT devices. Based on which, a knowledge-enhanced graph attention-based forecasting network is developed to predict the future behaviours of IoT devices. Anomalies, such as those caused by cyber-attacks in workflows, are detected by analyzing deviations from these predicted behaviours in conjunction with domain-specific knowledge. A case study is presented, along with extensive experiments conducted on publicly available datasets. Evaluation results demonstrate that KeAD outperforms the state-of-the-art techniques in terms of anomaly detection accuracy.

The usage of heterogeneous resources provisioned by edge nodes can be co-optimized through re-scheduling microservices. Current (re-)scheduling approaches typically treat the task of co-optimization as a single-objective optimization problem, which cannot address the issue of imbalanced usage of heterogeneous resources (e.g., CPU, memory, bandwidth) on a single edge node. More importantly, these approaches are inadequate in handling: (i) the adaptive co-optimization of heterogeneous resources, (ii) the fine-grained construction of microservice dependencies, and (iii) multi-step online microservice re-scheduling. To address these challenges, this article proposes a Dependency-aware Online Microservice re-Scheduling (DOMS) approach. DOMS formulates microservice re-scheduling as a multi-knapsack optimization problem and solves it using a Double Dueling Deep Q-Network (D3QN) with prioritized experience replay. Specifically, an adaptive heterogeneous resources balancing detection algorithm is developed, incorporating a dynamic detection threshold mechanism. A fine-grained microservice performance metrics dependency graph is constructed by capturing causal relationships to represent sequential execution dependency. Based on this graph, a microservice multi-step scheduling partition algorithm is devised. Extensive experiments are conducted upon publicly-available datasets, and evaluation results demonstrate that DOMS outperforms the state-of-the-art techniques with improvements of at least 1.85%, 6.45%, 0.56%, and 3.18% in terms of latency, energy consumption, balance degree, and throughput. These results highlight the effectiveness and superiority of DOMS in maintaining a balanced usage of heterogeneous resources and improving network throughput, while satisfying latency and energy consumption constraints.