The decentralized structure of federated learning exposes it to backdoor threats, especially dynamic black-box attacks. These attacks employ multiple evasion mechanisms within malicious updates, rendering many existing defenses ineffective. To address this threat, this paper proposes a three-stage defense algorithm. First, it performs neuron parameter analysis on PCA-reduced gradients to identify and remove decoy models injected by malicious clients. Second, a novel backdoor detection method is designed, where the central server generates random inputs to compare neuron activation differences across clients, thereby distinguishing between benign and backdoor gradients. Finally, Gaussian noise is added to disrupt the accuracy of the attacker’s feedback indicators. Compared to other defense algorithms, the proposed algorithm defends against each evasion module of dynamic black-box attacks while rendering the attacker’s feedback indicators ineffective. We evaluate the performance under different attack scenarios on four datasets, and the results show that the proposed algorithm effectively defends against various backdoor attacks and outperforms state-of-the-art methods.

Multimodal Large Language Models (MLLMs) integrate multimodal encoders with Large Language Models (LLMs) to overcome the limitations of text-only models. Traditional LLMs are deployed on high-performance cloud servers, but MLLMs, which process multimodal data, face high transmission latency and privacy risks when tasks are offloaded to the cloud. Intelligent edge computing is a promising solution for supporting such latency-sensitive and privacy-sensitive tasks. However, the heterogeneity of edge environments makes efficient MLLM inference challenging. In this work, we enhance MLLM inference efficiency in heterogeneous edge environments by decoupling MLLM into LLM and multimodal encoders, deploying the LLM on high-performance devices and the multimodal encoders on lower-capability devices. Additionally, we observe that processing MLLM tasks in edge environments involves numerous configuration parameters that impact inference speed and energy consumption in an unknown and possibly time-varying fashion. To address this challenge, we present an adaptive scheduling algorithm that assigns parameters to tasks or minimizing energy consumption while meeting maximum latency constraints. The results of extensive experimental trials demonstrate that the proposed approach consistently outperforms existing state-of-the-art methods, achieving significant improvements in both latency reduction and energy efficiency.

Large Language Models (LLMs) have demonstrated remarkable proficiency in generating text and producing fluent, succinct, and precise linguistic expressions. Limited battery life and computing power make it challenging to process LLM inference tasks in mobile devices. Intelligent edge computing brings the opportunity to help users process LLM inference tasks in real-time by offloading computations to nearby edge devices. However, due to the undetermined relationship between various task requirements and offloading configurations, inefficient offloading leads to unaffordable additional energy consumption, especially for intelligent tasks. This paper first investigates the energy consumption issue with different offloading configurations and task requirements in an intelligent edge testbed. According to the preliminary experiment results, we formulate the LLM offloading problem as a multi-armed bandit (MAB) problem and then use an upper confidence bound (UCB) bandit algorithm to find the energy-efficient offloading configurations. Extensive simulation results show that our approach enhanced the energy efficiency for offloading LLM inference tasks with different requirements in the intelligent edge environment.

Dynamically scaling Virtual Network Function (VNF) is essential for network operators to make their networks accommodate time-varying network traffic. However, current research only applies homogeneous VNF scaling or focuses on mathematical optimizations without implementations in a real-world environment. To build this gap, this paper presents HyScaler, a dynamic, hybrid VNF scaling system implemented in an open-source NFV platform. With HyScaler, we can build elastic Service Function Chains (SFCs) by scaling overloaded VNF Instances (VNFIs) across multiple CPU cores or physical servers. HyScaler implements this by (1) monitoring traffic loads for each deployed VNFI; (2) scaling overloaded VNFIs through all kinds of VNF scaling methods; and (3) placing and chaining the new VNFI using the Global VNF Placement and Chaining (GVPC) algorithm. Extensive experiments are conducted to validate the scalability of our HyScaler and the effectiveness of the GVPC algorithm. Compared with the original NFV platform, experimental results show that HyScaler can improve the performance of VNFIs by about 1.02 times.

Age of Information (AoI) is a newly emergent performance metric to quantify the freshness of data from destinations' perspectives. In this paper, we investigate and analyze AoI in the context of a multiple source updating system. In such a system, multiple loT devices continuously monitors physical environment and sends data to a remote destination for status updates through a Network Function Virtualization (NFV)-enabled network. Considering that the Virtual Network Function (VNF) placement can unnecessarily influence the AoI of the updates, we study the VNF placement problem in such a system. The problem is hence formulated as a mathematical optimization problem aiming to minimize the long-term average AoI of all updates received at the destination. To solve this prob-lem, we propose a Deep Reinforcement Learning (DRL)-based VNF placement approach called VNF-AoI, where a learning agent or decision-maker interacts with a system environment and consequently provides an optimal VNF placement policy according to the experience it has learned. Finally, we conduct extensive simulations to validate the effectiveness of our proposed approach. Numerical results clearly demonstrate that our VNF-AoI surpasses other two baseline algorithms by averagely 13.8 % higher acceptance ratio and 20.3 % lower average AoI at the destination.

Cloud computing is a scalable and efficient technology for providing different services. For better reconfigurability and other purposes, users build virtual networks in cloud environments. Since some applications bring heavy pressure to cloud datacenter networks, it is necessary to recognize and optimize virtual networks with different applications. In some cloud environments, cloud providers are not allowed to monitor user private information in cloud instances. Therefore, in this paper, we present a virtual network recognition and optimization method to improve quality-of-service (QoS) of cloud services. We first introduce a community detection method to recognize virtual networks from the cloud datacenter network. Then, we design a scheduling strategy by combining SDN-based network management and instance placement to improve the service-level agreements (SLA) fulfillment. Our experimental result shows that we can achieve a recognition accuracy as high as 80 percent to find out the virtual networks, and the scheduling strategy increases the number of SLA fulfilled virtual networks.

Graphics processing unit (GPU) accelerated processing performs significant efficiency in many multimedia applications. With the development of GPU cloud computing, more and more cloud providers focus on GPU-accelerated services. Since the high maintenance cost and different speedups for various applications, GPU-accelerated services still need a different pricing strategy. Thus, in this paper, we propose an optimal pricing strategy of GPU-accelerated multimedia processing services for maximizing the profits of both the cloud provider and users. We first analyze the revenues and costs of the cloud provider and users when users adopt GPU-accelerated multimedia processing services then state the profit functions of both the cloud provider and users. With a game theory based method, we find the optimal solutions of both the cloud provider's and users' profit functions. Finally, through large scale simulations, our pricing strategy brings higher profit to the cloud provider and users compared to the original pricing strategy of GPU cloud services.

Maintenance of communications during disasters is still a serious challenge due to power loss and equipment damage. An additional emergency communication usually needs a lot of investment and becomes a burden in normal. Since IoT devices become more and more important for human life, billions of connected devices become a great opportunity for building communications to resilient emergencies. In this paper, we present AWCT, a novel communication framework to support consistent communications with IoT devices in normal and emergency. In the AWCT framework, we design an orchestration of the low-power wide-area network (LPWAN) and ad hoc networks to build standby communication during emergencies. Meanwhile, we also optimize the cost of the AWCT framework by leveraging the standby time and battery capability. We test the AWCT framework with small testbed and extensive simulations. Evaluation results show that our work has great potential for building future disaster resilience IoT.

Internet of Things (IoT) becomes a very important area for providing various services on connected smart devices. For the isolation of different services in IoT, software-defined networking (SDN)-based virtual networks will be a scalable and flexible solution. However, in a large-scale IoT, as smart devices will move long distance between different positions, virtual network management becomes very difficult in providing network services. In this paper, we propose the LS-SDV, an efficient virtual network management framework in large-scale software-defined IoT (SDIoT). In this framework, we design a two-layer distributed control plane to manage devices and virtual networks in a large-scale environment. To the best of our knowledge, the LS-SDV is the first work to apply distributed control plane for virtual network management in softwarized networks. Moreover, based on the novel structure of the LS-SDV, we also provide a solution for network flow scheduling through network analysis. We evaluate the performance of our framework and virtual network management by extensive simulation and experiment in an open SDN framework.

Mobile crowdsensing becomes a promising technology for the emerging Internet of Things (IoT) applications in smart environments. Fog computing is enabling a new breed of IoT services, which is also a new opportunity for mobile crowdsensing. Thus, in this article, we introduce a framework enabling mobile crowdsensing in fog environments with a hierarchical scheduling strategy. We first introduce the crowdsensing framework that has a hierarchical structure to organize different resources. Since different positions and performance of fog nodes influence the quality of service (QoS) of IoT applications, we formulate a scheduling problem in the hierarchical fog structure and solve it by using a deep reinforcement learning–based strategy. From extensive simulation results, our solution outperforms other scheduling solutions for mobile crowdsensing in the given fog computing environment.

Radio frequency (RF) energy harvesting techniques are becoming a potential method to power battery-free wireless networks. In RF energy harvesting communications, energy cooperation enables shaping and optimization of the energy arrivals at the energy-receiving node to improve the overall system performance. In this article, we propose an energy cooperation scheme that enables energy cooperation in battery-free wireless networks with RF harvesting. We first study the battery-free wireless network with RF energy harvesting and then state the problem that optimizing the system performance with limited harvesting energy through new energy cooperation protocol. Finally, from the extensive simulation results, our energy cooperation protocol performs better than the original battery-free wireless network solution.

Vehicular ad-hoc networks (VANETs) will play an important role in next generation transportation systems, which is expected to be, cost-effective, and adaptable, making it ideal for providing network connection service to drivers and passengers. As a new paradigm of cloud computing, vehicular cloud computing (VCC) will improve scalability and flexibility of VANET services. However, traditional VANETs are hard to support VCC in the absence of network virtualization. In this paper, we propose a combined network virtualization scheme by introducing software defined networking (SDN) to VANETs. We combine two solutions of network isolation to enable network virtualization in software defined VANETs. To optimize the quality of service (QoS) of each virtual VANET, we also model the solution assignment as a non-cooperative game and find the Pareto efficient solution for fair assignment. From the experimental results, the new network virtualization scheme provides a better QoS than original solutions.

With the sharp growth of cloud services and their possible combinations, the scale of data center network traffic has an inevitable explosive increasing in recent years. Software defined network (SDN) provides a scalable and flexible structure to simplify network traffic management. It has been shown that Ternary Content Addressable Memory (TCAM) management plays an important role on the performance of SDN. However, previous literatures, in point of view on rule placement strategies, are still insufficient to provide high scalability for processing large flow sets with a limited TCAM size. So caching is a brand new method for TCAM management which can provide better performance than rule placement. In this paper, we propose FDRC, an efficient flow-driven rule caching algorithm to optimize the cache replacement in SDN-based networks. Different from the previous packet-driven caching algorithm, FDRC is characterized by trying to deal with the challenges of limited cache size constraint and unpredictable flows. In particular, we design a caching algorithm with low-complexity to achieve high cache hit ratio by prefetching and special replacement strategy for predictable and unpredictable flows, respectively. By conducting extensive simulations, we demonstrate that our proposed caching algorithm significantly outperforms FIFO and least recently used (LRU) algorithms under various network settings.

In cloud computing, companies usually use high-end storage systems to guarantee the I/O performance of virtual machines (VM). These storage systems cost a lot of energy for their high performance. In this paper, we propose an EEVS, a deduplication-based energy efficiency storage system for VM storage. We firstly investigate some VM image files with general operating systems. With the analysis result, we find there are many redundant data blocks that bring extra energy cost VM storage. Therefore, in the EEVS, we design an online-deduplication mechanism to reduce these redundant data without service interruption, while traditional deduplication technology is used for offline backup. Based on the system design, we implement an EEVS with the existing cloud platform. Since this mechanism needs considerable computing resources, we design a deduplication selection algorithm such that the storage energy consumption is minimized for a given set of VMs with limited resources for deduplication. Experiment results in a para-virtualization environments of the EEVS show that energy consumption is reduced by even up to 66% with negligible performance degradation.

With the increasing sensing and communication in cyber physical social system (CPSS), the data volume is growing much rapidly in recent years. Secure deduplication has attracted considerable interests of storage provider for data management efficiency and data privacy preserving. One of the most challenging issues in secure deduplication is how to manage data and the convergent key when users frequently update it. To solve this problem, D. Koo et al. use bilinear paring as the key method. However, bilinear paring requires high computation cost for implementations. In this paper, we propose a session-key-based convergent key management scheme, named SKC, to secure the dynamic update in the data deduplication. Specifically, each data owner in SKC can verify the correctness of the session key and dynamically change it with the data update. Furthermore, to enable group combination and remove the aid of gateway (GW), a convergent key sharing scheme, named CKS, is presented. Security analysis demonstrates that both SKC and CKS can protect the confidentiality of the data and the convergent key in the case of dynamic updates. The simulation results show that our SKC and CKS can significantly reduce computation complexity and communication during the data uploading phase.

With the surging popularity of smartphones and tablets, mobile network traffic has dramatically increased in recent years. Software defined network (SDN) provides a scalable and flexible structure to simplify network traffic management. It has been shown that rule placement plays an important role in the performance of SDN. However, since most existing work considers static network topologies of wired networks, it cannot be directly applied for mobile networks. In this paper, we propose MoRule, an efficient rule management scheme to optimize the rule placement for mobile users. To deal with the challenges of user mobility and rule capacity constraints of switches, we design a heuristic algorithm with low-complexity to minimize the rule space occupation while guaranteeing that the mobile traffic processed by local switches is no less than a threshold. By conducting extensive simulations, we demonstrate that our proposed algorithm significantly outperforms random solutions under various network settings.

Multicast is widely adopted by applications in cloud data centers. Due to the shortage of IP multicast groups, many efforts have been made to share multicast groups among multiple multicast sessions. However, existing group sharing schemes would lead to a serious privacy problem that some data will be exposed to receivers not belonging to the corresponding multicast session. In this paper, we propose a novel system called EPM for privacy-preserving multicast in cloud data centers. By enhancing a popular virtual machine manager Xen, it guarantees that the multicast packets received by each physical machine can be correctly forwarded only to virtual machines as destinations. Furthermore, we study a group assignment problem based on EPM to reduce redundant multicast traffic in cloud data centers, which will be solved by an efficient heuristic algorithm. We evaluate the effectiveness of EPM by implementing it on real systems, and conduct extensive experiments to show that the proposed algorithm can significantly reduce multicast traffic.

Software-defined network (SDN) is the next generation of networking architecture that is dynamic, manageable, cost-effective, and adaptable, making it ideal for the high-bandwidth, dynamic nature of today's applications. In SDN, network management is facilitated through software rather than low-level device configurations. However, the centralized control plane introduced by SDN imposes a great challenge for the network security. In this paper, we present a secure SDN structure, in which each device is managed by multiple controllers rather than a single one as in a traditional manner. It can resist Byzantine attacks on controllers and the communication links between controllers and SDN switches. Furthermore, we design a cost-efficient controller assignment algorithm to minimize the number of required controllers for a given set of switches. Extensive simulations have been conducted to show that our proposed algorithm significantly outperforms random algorithms.

Software-defined network (SDN) is the next generation of networking architecture that is dynamic, manageable, cost-effective, and adaptable, making it ideal for the high-bandwidth, dynamic nature of today’s applications. In SDN, network management is facilitated through software rather than low-level device configurations. However, the centralized control plane introduced by SDN imposes a great challenge for the network security. In this paper, we present a secure SDN structure, in which each device is managed by multiple controllers, not just a single as in a traditional manner, with the dynamic and isolated instance provided by the cloud. It can resist Byzantine attacks on controllers and the communication links between controllers and SDN switches. Furthermore, we study a controller minimization problem with security requirement and propose a cost-efficient controller assignment algorithm with a constant approximation ratio. From the experiment result, the secure SDN structure has little impact on the network latency, provide better security than general distributed controller, and the proposed algorithm performs higher efficiency than random assignment.