Modern deep learning models, such as Relation Graph Convolutional Network (RGCN), Sparse Convolutional Networks (SpConv), and Mixture of Experts Networks (MoE), are significantly dependent on the (gather-matmul-scatter) (abbreviated as (g-mm-s) ${ }_{\mathrm{s}}$) workload as their fundamental computational pattern. While existing works have made optimization attempts, several critical challenges remain unsolved, including domain-specific optimization migration, time-consuming exploration, and inefficient dataflow with dynamic inputs.To address these challenges, we introduce Efficient-GMS, a comprehensive framework that enhances ($\mathrm{g}-\mathrm{mm}-\mathrm{s})_{\text {s }}$ workload across diverse input scenarios. Our framework introduces (1) A Fusion-aware framework enabling cross-model optimization migration. We propose a comprehensive dataflow analysis that identifies shared computational patterns across models, enabling the development of four optimized dataflow patterns with vertical and horizontal fusion strategies. (2) Performance model-guided configuration space reduction. We develop a performance model to predict the relative execution efficiency across configurations, thereby reducing the search space and minimizing search time while ensuring optimal configuration selection. (3) Adaptive dataflow selection mechanism. We implement a lightweight heuristic model that dynamically selects optimal dataflow patterns based on the characteristics of the input and the hardware. Experimental results demonstrate that Efficient-GMS achieves significant performance gains, delivering an average end-to-end speedup of $1.46 \times$ in RGCN model, $1.32 \times$ in Sp-Conv-based model, and $1.15 \times$ in MoE model compared to state-of-the-art methods.

Mobile Edge Computing (MEC), as a real-time computing paradigm extended to the network edge, has been widely adopted. In recent years, Wireless Power Transfer-Aided Mobile Edge Computing (WPT-MEC) has garnered significant attention. However, it faces challenges in formulating effective offloading strategies and optimally allocating electrical energy resources. Existing solutions exhibit certain limitations, such as heuristic methods, incurring high computational complexity and struggle to adapt to dynamic environments. Although Deep Reinforcement Learning (DRL) overcomes the drawbacks of heuristic algorithms, it requires extensive time and training data. To address these issues, this paper proposes a DRL-Based Adaptive Offloading algorithm for WPT-MEC, termed as DRL-Based Adaptive Offloading (DRLAO) algorithm. This algorithm is able to dynamically adapt to environmental changes, make decisions rapidly, and adjust parameters in real-time. The DRLAO algorithm is comprised of three components: Augmented Deep Neural Network (AugDNN), Order-Preserving Quantization (KOQ) for addressing offloading decision-making, and Modi-fied Secant Method (MSM) for manipulating electrical energy resource allocation. The DRLAO algorithm achieves optimal performance of over 98% in different numbers of Wireless Edge Devices (WEDs) with lower CPU latency, and outperforms the baseline algorithm in terms of effectiveness and performance. In addition, it is able to quickly adapt and converge with minimal oscillation in dynamic environments. The source code is available at https://github.com/Aurora001226IDRLAO.

Internet of Things (IoT) revolutionizes data collection, especially in e-healthcare, where patients data from wearables and sensors improves medical services. However, IoT’s limitations in computing and storage require cloud outsourcing. Combining IoT with the cloud has potential but raises concerns about data security. Leveraging cloud storage presents an attractive solution for accommodating the substantial volume of data outsourced by IoT devices. As the outsourcing of real-time data to cloud storage becomes commonplace, the adoption of data auditing schemes emerges as a means to ensure data integrity. To curtail operational expenses, various deduplication techniques are commonly employed on outsourced data, effectively sidestepping redundant data and resulting in storage and bandwidth efficiencies. Although real-time data typically remains distinct due to its diverse origins, scenarios, such as data sharing or trading in data-driven services and datamarkets, can lead to data redundancy. Moreover, in order to fortify against any potential information leakage, encryption is implemented prior to deduplication. Convergent encryption (CE) stands as a prominent exemplar of this approach. Effectively integrating data auditing, deduplication, and encryption for wireless sensor devices is no trivial task. To efficiently and securely accommodate data while authenticating them through a heterogeneous framework, we present a novel remote data checking scheme, denoted as the VRDC scheme. This scheme empowers IoT data to be encrypted, updated, deduplicated, and audited, aligning with the imperatives of security, privacy, and efficiency. Through comprehensive security analysis, we establish that our VRDC scheme is fortified against potential threats. Our experimental findings highlight the efficiency of our approach in the realms of auditing, deduplication, and updates. Furthermore, the evidence highlights the potential for optimization within our scheme when compared to related works. This is achieved through the careful management of dynamic update scales within a file.

Industrial Robot Monitoring System (IRMS) is an important guarantee to maintain the normal operation of industrial robot systems. For IRMSs in the edge-cloud environment, live migration technology enables them to improve system resource utilization and reliability such as dynamic resource management or fault tolerance without interrupting monitoring services. Therefore, it is important to research the optimization of live migration for IRMS. For multi-container migration, parallel migration can reduce service downtime, serial migration can reduce pre-copy migration time, and hybrid migration with a reasonable serial-parallel relationship can combine the advantages of both. In this paper, we propose a multi-container migration architecture based on shared bandwidth, which considers the resource-constrained characteristics of the edge-cloud environment. Moreover, we present a multi-container hybrid migration planning model with the total migration time as the optimization objective, which uses a matrix representation of serial-parallel relationship. To solve this model, we develop a heuristic algorithm based on a hybrid Tabu-Evolutionary algorithm. The algorithm can find the dominant solution quickly by global search and improve the solution quality by subspace search. The experimental results show that the proposed algorithm can quickly give the hybrid migration strategy for a set of containers, effectively reducing the total migration time.

With the flourishing development of the Internet of Things (IoT) era, Edge Computing (EC) technology has garnered significant attention along with advancements in communication and IoT technologies. Effectively harnessing the computational resources in the edge environment has become a paramount concern. Accurate workload prediction is considered fundamental to optimizing the utilization of limited edge resources. However, most existing edge cloud load prediction approaches overlook the correlations among edge sites. Moreover, for cloud-native applications, user requests are typically handled by multiple containers. By evolving the work behavior of workload groups, it is possible to achieve a shift from focusing on individual containers to the collective, resulting in higher predictive accuracy. In this paper, the behavior of workload groups is analyzed through the reference to both static and dynamic container information. Leveraging the input data constructed based on the behavior of workload groups, an improved Sample Convolution and Interaction Network (SCINet) is employed for early multi-step prediction of edge container loads. Experimental validation on an edge cloud load dataset demonstrates the effectiveness of the proposed approach. Experimental results show that the proposed method can effectively improve the accuracy of edge cloud load prediction.