The convergence of networking and computing has become a critical architectural paradigm due to the increasing demand for services offering high availability and ultra-low latency, including immersive video and wide-area high-performance computing. However, increasing heterogeneity in computing resources and tighter coupling with algorithms have exacerbated the mismatch between resource supply and service demands. The imperative necessity for service-aware and resource-adaptive routing solutions is underscored by the fact that traditional location-based routing mechanisms neglect service semantics. Against this backdrop, the Computing Power Network (CPN) has emerged, integrating network and computing to enable dynamic, SLA-driven task orchestration. Although substantial progress has been achieved in task scheduling and orchestration in the existing work, the critical issue of service-level routing has not been completely addressed. To bridge this gap, we propose OSAF, a lightweight and backward-compatible service routing mechanism for CPN. OSAF introduced Service Tag (STag) as location-independent semantic identifier, which decouple service requests from host addresses. This enables users to access integrated services without being aware of resource locations. Our design of Service-Aware Routing (SAR) is based on OSAF. This routing algorithm utilizes reinforcement learning and multi-layer graph attention to model fine-grained resource dependencies and adapt to real-time service dynamics. Extensive evaluations of large-scale, dynamic service scenarios under CPN have shown that OSAF outperforms the current routing mechanisms in terms of SLA fulfillment, scalability, and load balancing.
Process models represent valuable experiential knowledge accumulated by enterprises throughout their development. The reuse of such knowledge can save managerial time and enhance the efficiency of process designers. Traditional process recommendation methods mostly focus on matching between process models themselves while neglecting the cross-modal matching problem between process texts and process models. Furthermore, enterprises accumulate process models in diverse forms, yet existing methods can only recommend a single form of process model (e.g., BPMN or Petri nets) and are unable to recommend multiple forms simultaneously. To address these challenges, this paper proposes a novel process recommendation network, referred to as ProcNet, which integrates large language models (LLMs) and graph convolutional networks (GCNs) for process recommendation. Specifically, (1) ProcNet uses LLMs to extract process activities and their relationships from requirement texts, constructs a directed graph, and semantically aligns it with the process activities in the database. Meanwhile, the process models in the database are also transformed into directed graphs; (2) ProcNet employs GCNs to obtain high-dimensional feature representations of each directed graph; (3) ProcNet calculates the top-k process models in the database most similar to the requirements using Euclidean distance and recommends them to the user. Extensive experiments demonstrate that ProcNet outperforms existing baselines, achieving remarkable results in recall, precision, F1-score, and mean average precision (MAP).
Cold starts are a big challenge in Function-as-a-Service (FaaS). While most solutions focus on optimizing the runtime environment and FaaS scheduling, they often overlook the impact of FaaS implementation on cold-start latency. Dynamic languages like Python, with their extensive dependency imports and dynamic building processes, can particularly suffer from longer startup times. We propose LightLoader, an approach to accelerating Python-based FaaS (i.e., PyFaaS) cold start by debloating them and optimizing their dynamic build. LightLoader rewrites PyFaaS by converting potentially unused functions to on-demand dynamic loading, thereby reducing loading time. It also delays third-party package loading until its first use instead of importing these all at the beginning. Notably, to ensure that these rewrites do not compromise the original PyFaaS’s functionalities, we utilize a large language model (LLM) to test and repair the optimized PyFaaS iteratively. We implemented a prototype on the popular open-source serverless platform OpenFaaS and conducted an extensive evaluation with PyFaaS of various sizes and functionalities. The experimental results show that LightLoader effectively accelerates PyFaaS cold starts, reducing end-to-end latency by 11.44
In service-oriented industrial supply chains, enterprises collaborate to share information while maintaining strict privacy requirements that prohibit raw financial data exchange between corporate groups. We propose a novel Granular Fuzzy Rule-Based Model based on Density Peaks Clustering (DPC-GFRBM) for privacy-preserving revenue prediction in industrial supply chain networks. DPC-GFRBM performs local clustering within each enterprise group to identify revenue patterns, then aggregates these results through collaborative analysis without revealing individual data. A fuzzy rule-based model maps the granulated features to revenue predictions, balancing privacy protection with predictive accuracy. Experimental results on public and real-world enterprise revenue datasets show that DPC-GFRBM consistently outperforms traditional fuzzy rule-based methods, achieving over 20
Optimizing route planning for Mobility-as-a-Service platforms, such as carpooling, involves solving combinatorial problems under precedence and multi-candidate constraints. Traditional methods struggle to balance real-time requirements and solution quality due to computational complexity. We introduce DeepQUBO, a quantum-compatible framework that reformulates the Precedence-Constrained Generalized Traveling Salesman Problem (GTSP-PC) into a sparse Quadratic Unconstrained Binary Optimization (QUBO) model. The framework employs auxiliary variables that decouple complex constraints, maintaining optimality while reducing QUBO elements by 51 × faster convergence than conventional approaches. Frustration analysis confirms DeepQUBO’s enhanced suitability for quantum optimization, achieving lower constraint-frustration ratios across benchmark datasets. Our work bridges service orchestration with quantum-ready optimization, offering practical value for large-scale transportation systems.
Business process simulation (BPS) has emerged as a crucial tool that offers a risk-free virtual environment to analyze, test, and optimize complex service compositions and orchestrations in cloud and edge computing environments. BPS enables the evaluation of alternative scenarios (a.k.a. “what-if” scenarios) by capturing the dynamic behavior of service interactions, including control-flow, task durations, and resource utilization. Several techniques to discover BPS models employ black-box predictors to characterize the run-time simulation aspects. The downside of these approaches is that the rules guiding the predictors cannot be provided in an intelligible form. This means that service architects and process analysts cannot modify or explore the decision-making logic that drives service orchestration and composition in these simulations. Moreover, these models are often deterministic and thus unable to capture uncertainty, which is essential to realistically simulate dynamic environments. This paper presents white-box predictors based on probabilistic decision trees, which are intelligible and easy to configure. The experiments show that white-box predictors can improve the simulation accuracy and variability, while being naturally intelligible.
In unmanned weighbridge Edge services, determining whether vehicles are correctly positioned on weighbridges from video streams efficiently at the Edge is a critical challenge. Traditional monocular deep models for vehicle localization are not lightweight and accurate enough for deploying on Edge devices. Meanwhile, existing inference scheduling methods often neglect multi-stream concurrency and effective collaboration among Edge devices. Therefore, we propose a monocular indirect vehicle localization method based on a lightweight keypoint detection model, optimized for NPUs through customized neural operators, together with a Multi-stream video Distributed Inference and Scheduling (MDIS) framework that integrates frame stitching, adaptive sampling, and weight-based load balancing. The proposals are evaluated on a real Edge device-based cluster, and experimental results show that our approach reduces parameters by 33.46 https://github.com/JiaQiChen-Lucas/vehicle_localization .
Process mining provides valuable insights into workflows based on recorded execution data. The registered event logs often contain sensitive information as they may bear personal details about working individuals or reveal organizations’ know-how and their operations’ history. Therefore, confidentiality and privacy requirements are high priority for process mining. In this paper, we present ProMiSe, a software framework and service that allows users to control information usage for automated process discovery, from input provision, through log processing, to output consultation. Usage control rules are expressed by means of policies. ProMiSe turns them into dedicated applications running within a Trusted Execution Environment (TEE) to enforce those rules. We put our solution to the test with real-world event logs to assess its computational resource consumption and service time.
We propose a novel framework for optimizing Quality of Service (QoS) fulfillment to meet drone providers’ satisfaction. The proposed framework leverages the provisioning of drone service needs in terms of on-time recharging to maximize providers’ satisfaction. A QoS fulfillment impact model is developed to assess the effect of inter-drone interference in meeting desired delivery time and energy consumption targets. A novel interference-aware algorithm is proposed for effective and efficient provisioning of drone service needs to fulfill their desired QoS targets. Experimental results conducted on real-world data validate the efficiency and effectiveness of the proposed approach.
QoS-based web service prediction is widely used in service computing but raises privacy concerns when collecting user data. Although local differential privacy (LDP) offers strong protection, its direct use in QoS prediction is hindered by attribute heterogeneity, which reduces data utility and prediction accuracy. To address this, we propose LDP-QWSP, an LDP-compliant framework that incorporates attribute-aware normalization and a perturbation mechanism favoring high-utility intervals. Experiments on real-world datasets confirm that LDP-QWSP achieves higher prediction accuracy than existing methods under comparable privacy budgets.
In uncertain environments, compliance assurance for business processes faces the intertwined challenges posed by diverse non-functional properties (NFPs) and complex gateways in the models. While solutions like user-defined metrics and constraints empower businesses to act independently, or Markov Decision Process (MDP) enhance statistical algorithm design, they address uncertainties from different views and have not been considered together to tackle a holistic, integrated business and system uncertainties. This paper proposes Process-aware MDP (P-MDP), a framework that unifies the two divergent view to optimize NFPs for business processes operating under uncertainty. We devise a gateway-aware, lazy-evaluation reward mechanism supporting the key stakeholders – business managers – to customize metrics, algorithms, and constraints, and apply them on business processes. Experiments with the WSDREAM benchmark dataset show that P-MDP outperforms the state-of-the-art (SOTA) method constraint-satisfied service composition MDP (CSSC-MDP) at various scales. Moreover, P-MDP demonstrates superior generality and scalability, enabling stakeholders to generate better execution plans for business processes in complex scenarios.
Blockchain technology has recently gained significant traction to support collaborative processes due to its ability to enable decentralization in environments that do not require a central authority. While there are numerous model-to-code approaches to facilitate the deployment and execution of processes on the blockchain, few have addressed the critical aspect of explicit modeling and designing blockchain-enabled processes. This paper introduces BPMN4BC, an extended set of notations designed to represent blockchain properties in process diagrams. These notations help distinguish on-chain from off-chain logic, preventing their overlap and accurately representing blockchain transactions. BPMN4BC enables business users and domain experts to capture blockchain-specific properties within their collaborative process models explicitly. We also introduce the BPMN4BC compiler that generates and deploys the corresponding Solidity code. To assess the consequences of different modeling choices, we compare the execution cost associated with two alternative deployment strategies, thereby demonstrating how modeling decisions influence blockchain properties such as gas usage.
Digital twins have become widely adopted in the building sector for operational monitoring and management; however, existing solutions rely heavily on users’ knowledge for decision-making and lack effective human-computer interaction mechanisms, limiting user engagement and actionable feedback. This paper introduces a human-computer interaction-enhanced digital twin visualization framework integrated with large language models (LLMs), designed to optimize occupants’ energy use by providing real-time insights into indoor environmental conditions and enhancing user engagement and actionable understanding through intuitive data exploration. By combining Building Information Modeling (BIM), Internet of Things (IoT) sensor networks, and LLMs, the framework enables users to interact with a 3D model of the building, monitor real-time data (e.g., temperature collected from edge devices), and receive environment adjustment suggestions to support decision-making. Compared to the conventional visualization platforms, the framework adds an additional interaction-feedback layer which combines predefined environmental threshold rules with a lightweight LLM to provide real-time prompts and context-aware suggestions to users. The 3D visualization environment is implemented using the Unity engine, enabling immersive interaction with building models and sensor data in real time. A locally hosted, lightweight LLM is embedded into the interface to support offline reasoning and privacy-preserving feedback generation. The system simulates the ability to detect and interpret abnormal building conditions and provide context-relevant responses, offering preliminary decision-support capabilities within the digital twin framework. Experimental results show that this system improves data comprehensibility, responsiveness, and user engagement, encouraging more sustainable occupant behavior. The integration of digital twins with lightweight interactive intelligence highlights a promising step toward adaptive and user-centered building operations. This paper also introduces our real implementation of the edge intelligence supported platform for an energy saving and net zero oriented building in collaboration with industry partners.
We propose a distributed trust information management framework for crowdsourced IoT services. The crowdsourced IoT service environment consists of distributed entities that store and manage trust information. Traditional trust management frameworks often assume the trustworthiness of these entities. However, they may tamper with trust data, making the system vulnerable to internal attacks. The rise of AI tools, such as ChatGPT, has further lowered the barrier for adversaries, enabling even low-skilled actors to manipulate trust information with alarming sophistication. AI-assisted tampering could severely compromise trust assessments and mislead IoT service users. To counter this emerging threat, we propose a novel AI-based tampering detection approach, complemented by a heuristic-based approach to identify tampered trust information. We conduct a series of experiments to evaluate the effectiveness of the proposed approaches. The results demonstrate more than 95
Software as a Service (SaaS) usually provides users with the ability to choose the features and usage limits they need – i.e. a configuration. Once a configuration is selected, it becomes a subscription. Despite enhancing value for customers, this model challenges providers, who must enforce that users operate within subscriptions that may change at any time – either by user choice or provider updates. This makes runtime subscription enforcement a self-adaptation challenge. A promising strategy to address this challenge is pricing-driven feature toggling, currently implemented only by Pricing4SaaS, to the best of our knowledge. This solution relies on two data sources: iPricings – machine-oriented representations of pricings; and subscription states – usage levels’ values of usage-limited features. However, Pricing4SaaS delegates subscription state management to the managed service, forcing providers to update subscription logic whenever a new usage limit is added/removed, which increases coupling and hinders scalability. This paper introduces three main contributions to address these limitations. First, we show that usage limits tend to increase as pricings evolve, reinforcing the need to simplify subscription management. Second, we introduce iSubscription, a machine-oriented model of subscription states compatible with iPricings. Third, we introduce SPACE, an independent service for pricing-driven self-adaptation that overcomes all state-of-the-art’s limitations.
Large Language Models (LLMs) are increasingly used to support intelligent agents in knowledge-intensive tasks. Yet, their application to business process discovery remains underexplored, especially in multi-agent settings requiring coordinated, multi-turn dialogue. This paper introduces an LLM-based Multi-Agent Systems (MAS) framework that uncovers business processes through structured conversational interactions. Using the Gaia methodology, we designed four MAS types (Monolithic, Duo, Manager, and Team) with varying levels of agent role specialization and coordination complexity. A custom, model-agnostic infrastructure is used to support the choreography-based collaboration of LLM-based agents. We evaluated our LLM-based MAS across three realistic business processes and three LLMs (GeminiPro 2.5, Mistral Large 2, DeepSeek-V3). GeminiPro consistently outperformed other models, reaching 100
To enhance resource utilization in data centers, it is common to co-locate latency-sensitive services and batch jobs on the same host, which leads to batch-to-online (B2O) and intra-batch (B-B) interference due to resource contention. Therefore, it is crucial to schedule jobs effectively to alleviate both types of interference. While most prior studies focus on B2O interference, B-B interference is often overlooked. By analyzing production data, we observe three insights that reveal the relationships between job behaviors and the two types of interference. We propose DIJS, a dual interference-aware job scheduling framework. DIJS predicts the resource usage patterns and resource sensitivity of new batch jobs. Based on this, it designs two quantizers using reinforcement learning (RL) and attention mechanisms to assess the impact of B2O and B-B interference. Experiments show that DIJS reduces service tail latency by about 41.0% and enhances job efficiency by up to 28.6% .
Deduplication and data sharing are two crucial technologies designed to enhance the storage and sharing efficiency of cloud service providers. However, existing encryption-based solutions face critical challenges that hinder their effectiveness. (1) Some existing encryption algorithms introduce randomness, which reduces deduplication efficiency and hampers data sharing. (2) Other algorithms attempt to balance deduplication and data sharing but rely on third-party authority servers. To address these challenges, we introduce the SDS-Cloud scheme, which reduces storage overhead for cloud services and promotes the sharing of data. This scheme operates independently of third-party authority servers for both deduplication and data sharing while defending against various types of attacks. Specifically, deduplication is achieved using an optimized message-locked encryption scheme. Unlike traditional message-locked encryption, which introduces randomness, our approach maintains deduplication efficiency while preventing brute-force dictionary attacks. Proxy re-encryption is employed to manage data access rights, enabling data sharing without granting the cloud service provider access to the actual data. Moreover, security analysis shows that the proposed scheme meets the required security standards. Furthermore, simulation experiments demonstrate that our scheme outperforms existing approaches in both repetitive data verification and permission management efficiency.
Quantum computing is an emerging technology that is increasingly accessed through cloud service platforms. However, the growing demand for quantum computing is constrained by the limited availability of resources, resulting in long waiting times for users and high execution costs, as well as under-utilization of quantum resources. To address this under-utilization, previous research has demonstrated that quantum task scheduling can effectively optimize the usage of quantum cloud platforms, combining multiple quantum circuits into a single execution task. This paper proposes an optimization method, QCRAFT AI Scheduler, to schedule quantum circuits making use of Artificial Intelligence algorithms. Specifically, this proposal is based on a Deep Learning model, which is capable of predicting the optimal scheduling of task queues, in order to use the maximum number of qubits of quantum computers. The validation results show that the proposal achieves 97.28
Internet of Things (IoT) systems are distributed collections of embedded systems that are connected over the Internet or similar networks. A key property of IoT systems is the need for high reliability throughout their lifespan under physical constraints, e.g., material properties, and natural degradation of system components. Simulation and digital twin approaches can quantify statistical assurances at design time and runtime respectively to verify reliability of IoT systems. However, existing approaches lack integration through a multi-paradigm modeling approach for seamless design time and runtime reliability verification, sole use of mature standards and protocols for interoperability and easy extension, and state-of-the-art statistical model checking (SMC) techniques for sound and efficient reliability verification. Consequently, we propose a service-oriented digital twin architecture and implement the architecture in a production-grade, interoperable, and radically distributed scale out process engine that integrates both design and runtime verification in a single model. We demonstrate our architecture and implementation for four IoT systems and report statistical reliability assurances through sound and efficient SMC.
