The integration of satellite communications with mobile edge computing (MEC) into space-air-ground integrated networks, known as satellite edge computing (SEC), has become a crucial research field for future communication systems to provide extensive global coverage services. This paper investigates the joint dependent task offloading and resource allocation problem for remote Internet-of-Things (IoT) applications within the SEC architecture. The proposed system leverages unmanned aerial vehicles (UAVs) as mobile access points and edge servers and utilizes low- earth orbit (LEO) satellites and ground stations as cloud computing resources. Multiple applications with dependent tasks from IoT devices (IoTDs) are modeled as directed acyclic graphs (DAGs). To address the challenges of reducing the system cost in UAV-assisted SEC, we first propose a one-to-many matching algorithm to associate IoTDs with UAVs. Then, a multi-application task sequence algorithm is devoted to merging the multiple DAGs and sorting the task order. Finally, a graph-aware asynchronous multi-agent reinforcement learning approach empowers the agents to autonomously discover optimal offloading and resource allocation strategies. Extensive simulations based on real-world datasets demonstrate the effectiveness of the proposed approach in minimizing the system costs while meeting application latency requirements, outperforming other benchmark algorithms.

The deployment of multimodal large language models (MLLMs) in edge-cloud networks faces critical challenges, including computational resource heterogeneity, memory bottlenecks, and bandwidth constraints. To address these issues, we propose EdgeSD, a novel framework that accelerates MLLM inference by integrating speculative decoding (SD) with edge-cloud collaboration. First, EdgeSD decouples the vision encoding and decoding processes of the draft MLLM across heterogeneous edge servers (ESs). This disaggregation architecture overcomes single-node memory constraints, enabling optimized resource utilization and high-resolution input processing. Second, to resolve the communication bottleneck and computational burden inherent in this distributed architecture, EdgeSD integrates a bandwidth-aware dynamic image token merging (ITM) method. Unlike general pruning techniques, this EdgeSD-specific ITM method focuses on minimizing inter-ES transmission latency for vision-decoding disaggregation while maintaining draft quality. Third, to optimize SD efficiency on consumer-grade ESs, EdgeSD employs an adaptive and scalable token tree structure solved using a parallel delta-stepping algorithm. This structure maximizes the number of accepted tokens under strict edge latency constraints. Extensive experiments on six multimodal datasets and five benchmarks with various MLLM pairs demonstrate that EdgeSD achieves substantial acceleration and throughput gains in edge-cloud collaboration scenarios using a lightweight draft MLLM, achieving 3.04-5.12x speedup compared to baseline methods.

The deployment of multimodal large language models (MLLMs) on resource-constrained mobile devices poses significant challenges due to their high computational demands. This paper introduces a novel two-timescale optimization framework for efficient MLLM inference in Edge-Cloud networks, addressing the problem of multi-timescale resource management by jointly optimizing slow-timescale MLLMs deployment decisions and fast-timescale batch scheduling, GPU resource allocation, and bandwidth allocation under dynamic network conditions and spatiotemporal request heterogeneity. Our key innovation is a hierarchical twin delayed deep deterministic policy gradient (HALTD3) algorithm that integrates attention mechanisms and long short-term memory networks to optimize slow-timescale MLLMs deployment and fast-timescale resource allocation, minimizing weighted system costs including deployment cost, end-to-end latency, and energy consumption, while meeting stringent quality-of-service requirements. Extensive experiments demonstrate that the HALTD3 algorithm substantially outperforms baseline methods in reducing system costs across diverse MLLM workloads and dynamic network scenarios, validating its effectiveness for practical edge-cloud collaborative inference.

Workflow scheduling plays a critical role in optimizing completion time and throughput in distributed cloud environments, leveraging the parallelism of heterogeneous computing resources. However, existing workflow scheduling algorithms often fall short due to heuristic limitations and the challenges in adaptability within heterogeneous settings, leading to suboptimal scheduling solutions. In this paper, we present a novel deep reinforcement learning (DRL) framework tailored for continuous workflow scheduling in heterogeneous environments. First, we propose an intelligent scheduler that updates the policy network through interactions with a multi-tenant environment, triggered by scheduling events. Next, the framework incorporates a Graph Attention Network (GAT) and a self-attention MultiLayer Perceptron (MLP) to preprocess the workflow topology and embed dynamic features of ready tasks and available processors into the state input at each scheduling step. Additionally, a k-dimensional tree-based k-nearest neighbors (kNN) algorithm is employed to map the output action vector to a pair of executed ready task and processor, facilitating the transition from continuous to discrete action spaces and addressing challenges associated with dynamic action spaces. Experimental results demonstrate that our method converges effectively in continuous workflow scheduling scenarios and significantly outperforms the best-known methods in terms of average makespan and load balancing efficiency.

Being an up-and-coming application scenario of mobile edge computing (MEC), the post-disaster rescue suffers multitudinous computing-intensive tasks but unstably guaranteed network connectivity. In rescue environments, quality of service (QoS), such as task execution delay, energy consumption and battery state of health (SoH), is of significant meaning. This paper studies a multi-user post-disaster MEC environment with unstable 5G communication, where device-to-device (D2D) link communication and dynamic voltage and frequency scaling (DVFS) are adopted to balance each user's requirement for task delay and energy consumption. A battery degradation evaluation approach to prolong battery lifetime is also presented. The distributed optimization problem is formulated into a mixed cooperative-competitive (MCC) multi-agent Markov decision process (MAMDP) and is tackled with recurrent multi-agent Proximal Policy Optimization (rMAPPO). Extensive simulations and comprehensive comparisons with other representative algorithms clearly demonstrate the effectiveness of the proposed rMAPPO-based offloading scheme.

Internet-of-Things (IoT) has played a critical role in developing sustainable smart cities and emerging numerous latency-sensitive IoT applications. Mobile edge computing (MEC) federation has the capability to incorporate a transparent resource management approach, which enables the sharing and utilization of MEC services from edge infrastructure providers (EIPs) and provides agile access services to mobile devices (MDs). In this paper, we investigate the joint optimization problem of computation offloading, task migration, and resource allocation in the MEC federation. The objective is to minimize the weighted sum of latency and energy consumption while maintaining load balancing under the constraint of the long-term migration cost budget of EIPs. To address the problem, we decompose it into two sub-problems: 1) the MDs clustering sub-problem and 2) the sub-problem of joint computation offloading, task migration, and resource allocation. Firstly, an MDs clustering matching (MDCM) algorithm is proposed to cluster the MDs in edge servers (ESs) according to the differences in channel gains. Afterward, the second sub-problem is simplified by the Lyapunov optimization technique, and then we propose a Transformer-based mobility prediction model and a decentralized deep deterministic policy gradient (DDPG)-based framework to solve it. Extensive simulation results demonstrate the cost-efficiency of the proposed algorithm.

Mobile edge computing (MEC) has emerged as a promising paradigm to enable computation-intensive and latency-sensitive mobile applications by offloading tasks to proximal edge servers. This paper proposes a novel federated reinforcement learning framework called Transformer-based Federated Soft Actor-Critic (TFSAC) to address a joint computation offloading and resource scheduling problem in a CPU-GPU heterogeneous MEC network while preserving privacy. Specifically, a graph attention network (GAT) extracts high-dimensional features from the task dependency graph. Rather than simply averaging weights, TFSAC applies transformer encoders to learn contextual relationships between agents and enable selective aggregation of relevant knowledge during federated model training to preserve agents’ privacy. Experiments on real-world trace data demonstrate TFSAC’s superiority over benchmarks in maximizing quality-of-service (QoS) across configurations.

Transformers are transforming the landscape of computer vision, especially for image-level recognition and instance-level detection tasks. Human-object interaction detection transformer (HOI-TR) is the first transformer-based end-to-end learning system for human-object interaction (HOI) detection; vision transformers build a simple multi-stage structure for multi-scale representation with single-scale patch and are the first patch-based transformer architecture for image-level recognition and instance-level detection. In this paper, we build a transformer-based multi-scale human-object interaction detector (MHOI), a novel method to integrate Vision and HOI detection Transformer, instead of directly incorporating two types of transformers, since the vision transformer lacks hierarchical architecture to handle the large variations in the scale of visual entities due to the single-scale patch partitioning. Specifically, MHOI embeds features of the same size (i.e., sequence length) with patches of variable scales simultaneously by utilizing overlapping convolutional patch embedding, then introduces an efficient transformer decoder that designs the query based on anchor points and essential auxiliary techniques to boost the HOI detection performance. Numerically, extensive experiments on several benchmarks demonstrate that our proposed framework outperforms prior existing methods coherently and achieves the impressive performance of 29.67 mAP on HICO-DET and 58.7 mAP on V-COCO, respectively.

Edge computing is an emerging promising computing paradigm that brings computation and storage resources to the network edge, hence significantly reducing the service latency and network traffic. In edge computing, many applications are composed of dependent tasks where the outputs of some are the inputs of others. How to offload these tasks to the network edge is a vital and challenging problem which aims to determine the placement of each running task in order to maximize the Quality-of-Service (QoS). Most of the existing studies either design heuristic algorithms that lack strong adaptivity or learning-based methods but without considering the intrinsic task dependency. Different from the existing work, we propose an intelligent task offloading scheme leveraging off-policy reinforcement learning empowered by a Sequence-to-Sequence (S2S) neural network, where the dependent tasks are represented by a Directed Acyclic Graph (DAG). To improve the training efficiency, we combine a specific off-policy policy gradient algorithm with a clipped surrogate objective. We then conduct extensive simulation experiments using heterogeneous applications modelled by synthetic DAGs. The results demonstrate that: 1) our method converges fast and steadily in training; 2) it outperforms the existing methods and approximates the optimal solution in latency and energy consumption under various scenarios.

Vehicular edge computing (VEC) is a new computing paradigm that has great potential to enhance the capability of vehicle terminals (VTs) to support resource-hungry in-vehicle applications with low latency and high energy efficiency. In this article, we investigate an important computation offloading scheduling problem in a typical VEC scenario, where a VT traveling along an expressway intends to schedule its tasks waiting in the queue to minimize the long-term cost in terms of a tradeoff between task latency and energy consumption. Due to diverse task characteristics, dynamic wireless environment, and frequent handover events caused by vehicle movements, an optimal solution should take into account both where to schedule (i.e., local computation or offloading) and when to schedule (i.e., the order and time for execution) each task. To solve such a complicated stochastic optimization problem, we model it by a carefully designed Markov decision process (MDP) and resort to deep reinforcement learning (DRL) to deal with the enormous state space. Our DRL implementation is designed based on the state-of-the-art proximal policy optimization (PPO) algorithm. A parameter-shared network architecture combined with a convolutional neural network (CNN) is utilized to approximate both policy and value function, which can effectively extract representative features. A series of adjustments to the state and reward representations are taken to further improve the training efficiency. Extensive simulation experiments and comprehensive comparisons with six known baseline algorithms and their heuristic combinations clearly demonstrate the advantages of the proposed DRL-based offloading scheduling method.

Inspired by mobile edge computing (MEC), vehicular edge computing (VEC) enables vehicle terminals to support resource-hungry on-vehicle applications with significantly lower latency and less energy consumption. In this paper, we investigate the computation offloading problem in a typical VEC scenario, where a vehicle offloads its computation tasks to the VEC servers deployed in the road side unit (RSU) to minimize its long-term user cost. The mobility of the vehicle coupled with the high dynamics of the environment makes the problem particularly difficult. To tackle this challenge, a deep reinforcement learning (DRL) based offloading method is proposed, which approximates the offloading policy (OP) by a deep neural network (DNN) and trains the DNN with the proximal policy optimization (PPO) algorithm without a priori knowledge of the environment dynamics. Extensive simulation experiments and comprehensive comparison with six baseline algorithms demonstrate that it can achieve the lowest user cost in most cases.

Summary Metadata operations have become dominant file operations in the storage systems. In the scenarios of read‐more and write‐less of massive small files, the current distributed file systems suffer from the unsatisfying performance and scalability of metadata service because of random disk I/O during metadata operations. In this paper, a highly efficient metadata management architecture for cloud storage systems is proposed. The cluster design significantly improves the scalability of the system. A concept of the disk I/O channel is introduced, which is an independent data storage pipe by binding an independent CPU‐core to each physical disk. In addition, a multi‐channel fast key‐value storage engine is proposed to provide the extremely efficient performance for the underlying storage service, which takes full advantages of multi‐core processors and parallel disks I/O. Besides, a new dynamic load‐balancing strategy is proposed to reduce load thrashing and improve the precision of rebalancing among the clusters. Performance measurements under a variety of benchmarks show that the metadata management is capable of handling the massive small files storage and the performance is improved significantly compared to the existing solutions. Copyright © 2016 John Wiley & Sons, Ltd.

Summary With the rapid development of intelligent transportation technology, high‐definition data storage, and processing of massive amounts of video surveillance have become key issues. When thousands of high‐bit‐rate video streaming concurrently writes, disk I/O throughput becomes a bottleneck. In addition, this leads to serious energy consumption and disk abrasion. To solve these problems, a new distributed file system for high concurrent and high‐bit‐rate writing is designed. It combines an optimized data storage model, efficient metadata management, and exquisite disk schedule mechanism. The optimized data storage model uses a file pre‐allocation strategy and multiple‐stream input modulating technology to convert the randomly concurrent writes on the disk into sequential writes; the metadata management provides an efficient means of retrieving the specified data; and the dual‐partition schedule mechanism can ensure the disk stability with less abrasion. Through this distributed file system, the disk I/O throughput can be saturated in a high concurrent writing environment. The performance evaluation results demonstrate that the I/O throughput of a normal 7200RPM SATA III disk in our scheme can be stabilized at 150MB/s, easily to support 300 concurrent high‐definition video streams (4Mbit/s each). The distributed file system with eight commodity servers can afford the ability of supporting 8000 high‐definition video streams concurrently writing, which is far greater than the existing video surveillance storage solutions. Copyright © 2015 John Wiley & Sons, Ltd.