面向移动边缘计算中多应用服务的虚拟机部署算法

移动边缘计算(MEC)通过在用户近端以虚拟机(VM)形式部署应用服务,能有效降低服务响应延迟并减少核心网络数据流量。然而,当前MEC中虚拟机部署的大多数研究尚未具体考虑用户对多种应用服务的需求。因此,该文针对MEC中多应用服务的虚拟机部署问题,提出两种启发式算法,即基于适应度的启发式部署算法(FHPA)和基于分治的启发式部署算法(DCBHPA),通过在边缘网络中配置支持多种应用服务的虚拟机来最大限度地减少网络中的数据流量。FHPA和DCBHPA分别基于边缘服务器的网络连接特征和用户对应用请求的差异性,定义了不同的适应度计算模型。在此基础上,通过子问题划分机制实现VM配置。仿真结果表明,相比于基准算法,所提算法能更好地控制系统数据流量,有效地提高边缘网络服务资源的利用率。     

一种车载服务的快速深度Q学习网络边云迁移策略

智能网联交通系统中车载用户的高速移动，不可避免地造成了数据在边缘服务器之间频繁迁移，产生了额外的通信回传时延，对边缘服务器的实时计算服务带来了巨大的挑战。为此，该文提出一种基于车辆运动轨迹的快速深度Q学习网络(DQN-TP)边云迁移策略，实现数据迁移的离线评估和在线决策。车载决策神经网络实时获取接入的边缘服务器网络状态和通信回传时延，根据车辆的运动轨迹进行虚拟机或任务迁移的决策，同时将实时的决策信息和获取的边缘服务器网络状态信息发送到云端的经验回放池中；评估神经网络在云端读取经验回放池中的相关信息进行网络参数的优化训练，定时更新车载决策神经网络的权值，实现在线决策的优化。最后仿真验证了所提算法与虚拟机迁移算法和任务迁移算法相比能有效地降低时延。     

移动边缘网络中基于用户移动的服务功能链迁移策略

移动边缘计算(Mobile Edge Computing,MEC)通过在网络边缘部署服务器,提供计算和存储资源,可为用户提供超低时延和高带宽业务。网络功能虚拟化(Network Function Virtualization,NFV)与MEC技术相结合,可在MEC服务器上提供服务功能链(Service Function Chain,SFC),提升用户的业务体验。为了保证移动用户的服务质量,需要在用户跨基站移动时将SFC迁移到合适的边缘服务器上。主要以最小化用户服务的端到端时延和运行成本为目标,提出了MEC网络中具有资源容量约束的SFC迁移策略,以实现移动用户业务的无缝迁移。仿真结果表明,与现有方案相比,该策略具有更好的有效性和高效性。     

Dynamic service migration strategy based on MDP model with multiple parameter in vehicular edge network

To handle with the service interruption caused by vehicles’ mobility and limited service coverage of edge servers,a dynamic service migration algorithm based on multi-parameters Markov decision process (MDP) model was put forward for vehicular edge network,which was called as dynamic service migration algorithm based on multiple parameter (DSMMP).Combining delay,bandwidth,server capacity with vehicle motion information,DSMMP constructed a multi-parameters MDP revenue function to remedy the deficiency of distance-based schemes.By using vehicle motion and delay constraints,a candidate server set with several candidate servers was defined,and migration decision through long-term Bellman revenue values was made.In order to improve the dynamic adaptability of the proposed algorithm,the weight values were calculated and updated by leveraging historical information.Simulation results show that our strategy has a good performance in terms of delay,packet loss ratio and service migration times.     

移动性感知的边缘服务迁移策略

针对移动边缘计算网络中由于用户位置动态变化而导致边缘服务器间负载不均衡、用户服务质量降低的问题,提出了一种移动性感知的边缘服务迁移算法。首先,以最小化用户服务请求感知时延为目标,将优化问题建模为混合整数非线性规划问题。其次,基于Lyapunov优化方法将时延优化问题解耦为边缘服务迁移子问题与无线接入子问题。再次,提出快速边缘决策算法求解出给定无线接入策略情况下最优的资源分配与边缘服务迁移方案。最后,提出异步最佳响应算法迭代出最优无线接入策略。仿真结果表明,与现有服务迁移策略相比较,所提算法能够在保证服务迁移成本稳定的情况下降低用户服务请求的感知时延。     

Deep reinforcement learning-based joint task offloading and bandwidth allocation for multi-user mobile edge computing

The rapid growth of mobile internet services has yielded a variety of computation-intensive applications such as virtual/augmented reality. Mobile Edge Computing (MEC), which enables mobile terminals to offload computation tasks to servers located at the edge of the cellular networks, has been considered as an efficient approach to relieve the heavy computational burdens and realize an efficient computation offloading. Driven by the consequent requirement for proper resource allocations for computation offloading via MEC, in this paper, we propose a Deep-Q Network (DQN) based task offloading and resource allocation algorithm for the MEC. Specifically, we consider a MEC system in which every mobile terminal has multiple tasks offloaded to the edge server and design a joint task offloading decision and bandwidth allocation optimization to minimize the overall offloading cost in terms of energy cost, computation cost, and delay cost. Although the proposed optimization problem is a mixed integer nonlinear programming in nature, we exploit an emerging DQN technique to solve it. Extensive numerical results show that our proposed DQN-based approach can achieve the near-optimal performance.     

Workload Prediction Based Virtual Machine Migration and Optimal Switching Strategy for Cloud Power Management

Virtual Machine (VM) Migration has been popular nowadays, as it helps to balance the load effectively. Various VM migration-based approaches are modeled for better VM placement but remain the challenge because of inappropriate load balancing. Thus, workload prediction-based VM migration is introduced to improve the energy efficiency of the system. Importantly, load prediction is very important to enhance resource allocation and utilization. Chaotic Fruitfly Rider Neural Network is devised by combining Rider neural network and chaotic Fruitfly optimization algorithm to predict load. Moreover, the fitness for predicting the load is based on old-time load, resource constraint, and network parameters. Once the load is predicted, the power optimization is performed using VM migration and optimal switching strategy. When the load is found overloaded, the VM migration is performed using the proposed Harris Hawks spider monkey optimization (HHSMO). Thus, the optimal finding of VM for executing the removed task is found out using the proposed HHSMO. The fitness function utilized for the VM migration is based on power, load, and resource parameter. If the load predicted is underloaded, the optimal switch ON/OFF is done optimally by switch ON/OFF the servers using the proposed HHSMO algorithm. Through the migration and switching strategy, the power consumption is optimized. The performance of the proposed model is evaluated in terms of power consumption, load, and resource utilization. The proposed HHSMO achieves the minimal power consumption of 0.0181, the minimal load of 0.002, and the minimal resource utilization of 0.0376.

     

Congestion resulting from increased capacity in single-serverqueueing networks

In many networks for traffic and telecommunications, minimizing delays from entry to exit is a major concern of users. In user-optimal routing, each user chooses a path to minimize delay from entry to exit, given the existing paths chosen by all other users. Under user-optimal routing, at equilibrium all users experience the same delay. Many networks, especially data networks, are commonly modeled as networks of single-server queues. We report examples of single-server queueing networks with user-optimal routing in which adding servers or increasing the capacity of existing servers worsens the delay experienced by all users     

A genetic‐based decision algorithm for multisite computation offloading in mobile cloud computing

Mobile cloud computing is a promising approach to improve the mobile device's efficiency in terms of energy consumption and execution time. In this context, mobile devices can offload the computation‐intensive parts of their applications to powerful cloud servers. However, they should decide what computation‐intensive parts are appropriate for offloading to be beneficial instead of local execution on the mobile device. Moreover, in the real world, different types of clouds/servers with heterogeneous processing speeds are available that should be considered for offloading. Because making offloading decision in multisite context is an NP‐complete, obtaining an optimal solution is time consuming. Hence, we use a near optimal decision algorithm to find the best‐possible partitioning for offloading to multisite clouds/servers. We use a genetic algorithm and adjust it for multisite offloading problem. Also, genetic operators are modified to reduce the ineffective solutions and hence obtain the best‐possible solutions in a reasonable time. We evaluated the efficiency of the proposed method using graphs of real mobile applications in simulation experiments. The evaluation results demonstrate that our proposal outperforms other counterparts in terms of energy consumption, execution time, and weighted cost model.     

超密集网络中基于移动边缘计算的任务卸载和资源优化

移动边缘计算(MEC)通过在无线网络边缘为用户提供计算能力,来提高用户的体验质量。然而,MEC的计算卸载仍面临着许多问题。该文针对超密集组网(UDN)的MEC场景下的计算卸载,考虑系统总能耗,提出卸载决策和资源分配的联合优化问题。首先采用坐标下降法制定了卸载决定的优化方案。同时,在满足用户时延约束下采用基于改进的匈牙利算法和贪婪算法来进行子信道分配。然后,将能耗最小化问题转化为功率最小化问题,并将其转化为一个凸优化问题得到用户最优的发送功率。仿真结果表明,所提出的卸载方案可以在满足用户不同时延的要求下最小化系统能耗,有效地提升了系统性能。     

车载边缘计算中多任务部分卸载方案研究

现有车载应用设备对时延有更严苛的要求,车载边缘计算(VEC)能够充分利用网络边缘设备,如路边单元(RSU)进行协作处理,可有效地降低时延。现有研究多假设RSU计算资源充足,可提供无限的服务,但实际其计算资源会随着所需处理任务数量的增加而受限,对时延敏感的车载应用造成限制。该文针对此问题,提出一种车载边缘计算中多任务部分卸载方案,该方案在充分利用RSU的计算资源条件下,考虑邻近车辆的剩余可用计算资源,以最小化总任务处理时延。首先在时延限制和资源约束下分配各任务在本地、RSU和邻近车辆的最优卸载决策变量比例,其次以最小处理时延为目的在一跳通信范围内选择合适的空闲车辆作为处理部分任务的邻近车辆。仿真结果表明所提车载边缘计算中多任务部分卸载方案相较现有方案能较好地降低时延。     

RAD‐EI: A routing attack detection scheme for edge‐based Internet of Things environment

Internet of Things (IoT) offers various types of application services in different domains, such as “smart infrastructure, health‐care, critical infrastructure, and intelligent transportation system.” The name edge computing signifies a corner or edge in a network at which traffic enters or exits from the network. In edge computing, the data analysis task happens very close to the IoT smart sensors and devices. Edge computing can also speed up the analysis process, which allows decision makers to take action within a short duration of time. However, edge‐based IoT environment has several security and privacy issues similar to those for the cloud‐based IoT environment. Various types of attacks, such as “replay, man‐in‐the middle, impersonation, password guessing, routing attack, and other denial of service attacks” may be possible in edge‐based IoT environment. The routing attacker nodes have the capability to deviate and disrupt the normal flow of traffic. These malicious nodes do not send packets (messages) to the edge node and only send packets to its neighbor collaborator attacker nodes. Therefore, in the presence of such kind of routing attack, edge node does not get the information or sometimes it gets the partial information. This further affects the overall performance of communication of edge‐based IoT environment. In the presence of such an attack, the “throughput of the network” decreases, “end‐to‐end delay” increases, “packet delivery ratio” decreases, and other parameters also get affected. Consequently, it is important to provide solution for such kind of attack. In this paper, we design an intrusion detection scheme for the detection of routing attack in edge‐based IoT environment called as RAD‐EI. We simulate RAD‐EI using the widely used “NS2 simulator” to measure different network parameters. Furthermore, we provide the security analysis of RAD‐EI to prove its resilience against routing attacks. RAD‐EI accomplishes around 95.0% “detection rate” and 1.23% “false positive rate” that are notably better than other related existing schemes. In addition, RAD‐EI is efficient in terms of computation and communication costs. As a result, RAD‐EI is a good match for some critical and sensitive applications, such as smart security and surveillance system.     

数据新鲜度驱动的协作式无人机联邦学习智能决策优化研究

联邦学习是6G关键技术之一,其可以在保护数据隐私的前提下,利用跨设备的数据训练一个可用且安全的共享模型。然而,大部分终端设备由于处理能力有限,无法支持复杂的机器学习模型训练过程。在异构网络融合环境下移动边缘计算(MEC)框架中,多个无人机(UAVs)作为空中边缘服务器以协作的方式灵活地在目标区域内移动,并且及时收集新鲜数据进行联邦学习本地训练以确保数据学习的实时性。该文综合考虑数据新鲜程度、通信代价和模型质量等多个因素,对无人机飞行轨迹、与终端设备的通信决策以及无人机之间的协同工作方式进行综合优化。进一步,该文使用基于优先级的可分解多智能体深度强化学习算法解决多无人机联邦学习的连续在线决策问题,以实现高效的协作和控制。通过采用多个真实数据集进行仿真实验,仿真结果验证了所提出的算法在不同的数据分布以及快速变化的动态环境下都能取得优越的性能。     

Offloading framework for computation service in the edge cloud and core cloud: A case study for face recognition

A fast rate of progress has allowed the proliferation of smartphones and eased their extensive presence in people's daily life. However, low processing speed and limited battery capacity have hindered improvements in the smartphone's computational capabilities. Offloading computational tasks to the cloud could solve this problem by enabling users to access these services over the Internet. Edge cloud computing has been recognized as an emerging field within the cloud computing paradigm, where computation servers are situated at the edge of the Internet to reduce network delay and traffic. Nevertheless, offloading tasks to the cloud is not always beneficial due to variable network conditions and increased processing costs. In this paper, a deep reinforcement learning-based offloading framework has been presented that provides smartphones with the ability to make decisions for local processing in the smartphone or to offload processing tasks to the cloud (edge and/or core). Thus, a smartphone can minimize the combination of the processing time, energy consumption, and monetary cost and maximize the accuracy of face recognition as well. Simulation results under synthetic scenarios show that the proposed offloading framework can effectively adapt to the dynamic cloud computing and networking environment.     

PAVM: a framework for policy‐aware virtual machine management

The problem of efficient placement of virtual machines (VMs) in cloud computing infrastructure is well studied in the literature. VM placement decision involves selecting a physical machine in the data center to host a specific VM. This decision could play a pivotal role in yielding high efficiency for both the cloud and its users. Also, reallocation of VMs could be performed through migrations to achieve goals like higher server consolidation or power saving. VM placement and reallocation decisions may consider affinities such as memory sharing, CPU processing, disk sharing, and network bandwidth requirements between VMs defined in multiple dimensions. Considering the NP‐hard complexity associated with computing an optimal solution for this VM placement decision problem, existing research employs heuristic‐based techniques to compute an efficient solution. However, most of these approaches are restricted to only a single attribute at a time. That is, a given technique of using heuristics to compute VM placement considers only a single attribute, while completely ignoring the impact of other dimensions of placing VMs. While this approach may improve the efficiency with respect to the affinity attribute in consideration, it may yield degraded performance with respect to other affinities. In addition, the criteria for determining VM‐placement efficiency may vary for different applications. Hence, the overall goal of achieving VM placement efficiency becomes difficult and challenging. We are motivated by this challenging problem of efficient VM placement and propose policy‐aware virtual machine management (PAVM), a generic framework that can be used for efficient VM management in a cloud computing platform based on the service provider‐defined policies to achieve the desired system‐wide goals. This involves efficient means to profile different VM affinities and to use profiled information effectively by intelligent and efficient VM migrations at run time considering multiple attributes at a time. By conducting extensive evaluation through simulation and real experiments that involve VM affinities on the basis of network and memory, we confirmed that the PAVM architecture is capable of improving the efficiency of a cloud system. We elaborate the architecture of a PAVM system, describe its implementation, and present details of our experiments. Copyright © 2016 John Wiley & Sons, Ltd.     

基于自然梯度Actor-Critic强化学习的卫星边缘网络服务功能链部署方法

鉴于低轨卫星网络的高动态性和空间环境的复杂性,如何提供在线的快速服务功能链(SFC)部署方法,成为低轨卫星边缘网络中亟待解决的问题。综合考虑节点和链路容量等约束以及服务迁移等切换代价,针对部署多接入边缘计算(MEC)服务器的低轨卫星网络,该文提出一种基于自然梯度参与者-评价者(Actor-Critic)强化学习架构的SFC在线部署方法。首先,针对低轨卫星网络的环境高动态性,对实时容量约束和迁移代价进行建模;其次,引入马尔可夫决策过程(MDP),综合考虑服务迁移和卫星坐标等因素,描述低轨卫星网络的状态转移过程;最后,提出一种基于自然梯度的在线SFC部署强化学习方法,不同于标准梯度,自然梯度法进行模型层面的更新,以避免神经网络的训练陷入局部最优解。仿真结果表明,该文方法可逼近全局最优解,并在端到端时延性能上优于基于标准梯度的强化学习部署方法。     

移动边缘计算中基于联邦学习的视频请求预测和协作缓存策略

随着互联网社交平台的崛起和移动智能终端设备的普及,自媒体短视频、直播等视频业务蓬勃发展,人们对高质量视频服务的需求也急剧上升。与此同时,连接到核心网络的大量智能设备增加了回程链路的负载,传统的云计算难以满足用户对视频服务的低延迟要求。移动边缘计算(MEC)通过在网络边缘部署具有计算和存储能力的边缘节点,通过在更靠近用户的边缘侧提高计算和存储服务,降低了数据传输时延进而缓解了网络阻塞。因此,基于MEC架构,该文充分利用网络边缘资源,提出了基于联邦学习的视频请求预测和视频协作缓存策略。通过利用多个边缘节点对提出的深度请求预测模型(DRPN)视频请求预测模型进行联邦训练,预测视频未来的请求情况,然后量化缓存内容所带来的时延收益并协作地以最大化该时延收益为目的进行缓存决策。该文分析了真实数据集MovieLens,模拟了视频请求缓存场景并进行实验。仿真结果表明,相比于其他策略,所提策略不仅能有效降低用户等待时延,在有限的缓存空间中提高内容多样性,从而提高缓存命中率,降低缓存成本,还能降低整个系统的通信成本。     

WSN中基于强化学习的能效优化任务处理机制

以提高无线传感器网络中任务处理的能效为目标,提出了一种近似最优化的任务处理机制,无线传感器节点可根据任务缓存区的任务数量、信道条件,动态地实现任务向边缘服务器的卸载以及本地处理.将任务处理机制建模为马尔可夫决策过程,因为无线传感器节点不知道此过程的状态转移概率,所以采用A3C算法以实现在环境参数未知情况下的探索和学习,...     

SEAL2: An SDN‐enabled all‐Layer2 packet forwarding network architecture for multitenant datacenter networks

This paper presents the design and development of a new network virtualization scheme to support multitenant datacenter networking (MT‐DCN) based on software‐defined networking (SDN) technologies. Effective multitenancy supports are essential and challenging for datacenter networking designs. In this study, we propose a new network virtualization architecture framework for efficient packet forwarding in MT‐DCN. Traditionally, an internet host uses IP addresses for both host identification and location information, which causes mobile IP problems whenever the host is moved from one IP subnet to another. Unfortunately, virtual machine (VM) mobility is inevitable for cloud computing in datacenters for reasons such as server consolidation and network traffic flow optimization. To solve the problems, we decouple VM identification and location information with two independent values neither by IP addresses. We redefine the semantics of Ethernet MAC address to embed tenant ID information to the MAC address field without violating its original functionality. We also replace traditional Layer2/Layer3 two‐stage routing schemes (MAC/IP) with an all‐Layer2 packet forwarding mechanism that combines MAC addresses (for VM identification and forwarding in local server groups under an edge switch gateway) and multiprotocol label switching (MPLS) labels (for packet transportation between edge switch gateways across the core label switching network connecting all the edge gateways). To accommodate conventional IP packet architecture in a multitenant environment, SDN (OpenFlow) technology is used to handle all this complex network traffics. We verified the design concepts by a simple system prototype in which all the major system components were implemented. Based on the prototype system, we evaluated packet forwarding efficiency under the proposed network architecture and compared it with conventional IP subnet routing approaches. We also evaluated the incurred packet processing overhead caused by each of the packet routing components.     

Games on networks

It is suggested that user selection of service provision on a network of heterogeneous servers be modeled by competitive replicator dynamics. User agents make their requirements known to network agents, which can bid for their execution. The bids initiate a negotiation process between the two groups of agents, at the end of which the user agents make a decision and map their requirements onto a selected set of network servers. The selection process is based on competitive replicator dynamics and attempts to maximize the respective utilities of both agent groups. Various issues resulting from the nonlinearity of the dynamics, such as stability and adaptability, are discussed in some detail. The results of some simulations are presented