基于Kubernetes的多云网络成本优化模型

以Kubernetes为代表的云原生编排系统在多云环境中被云租户广泛使用,随之而来的网络观测性问题愈发突出,跨云跨地区的网络流量成本尤为突出。在Kubernetes中引入扩展的伯克利数据包过滤器（extended Berkeleypacketfilter,eBPF）技术采集操作系统内核态的网络数据特征解决网络观测问题,随后将网络数据特征建模为二次分配问题（quadratic assignment problem,QAP）,使用启发式搜索与随机搜索组合的方法在实时计算的场景下求得最佳近优解。此模型在网络资源成本优化中优于Kubernetes原生调度器中仅基于计算资源的调度策略,在可控范围内增加了调度链路的复杂度,有效降低了多云多地区部署环境中的网络资源成本。     

异构备份式的虚拟网映射方法研究

在云计算和数据中心环境中,底层单个物理服务器的失效将对上层虚拟网络的服务性能造成很大的影响,现有利用冗余备份的方法能够在一定程度上降低底层物理设备失效带来的影响,但未考虑到物理服务器的同构性所带来的问题,为此,该文提出一种异构备份式的虚拟网映射方法。首先,只对关键的虚拟机进行冗余备份,降低备份资源的开销;然后,确保提供备份虚拟机的物理服务器与原物理服务器的系统类型的异构性,提高虚拟网的弹性能力;最后,以最小化链路资源开销作为虚拟网的映射目标,进一步降低备份资源的开销。实验表明,该方法在保证虚拟网络映射性能的前提下,能够大大提高虚拟网络的弹性能力。     

业务需求感知的虚拟数据中心映射方法研究(英文)

Virtualization is a common technology for resource sharing in data center.To make efficient use of data center resources,the key challenge is to map customer demands(modeled as virtual data center,VDC) to the physical data center effectively.In this paper,we focus on this problem.Distinct with previous works,our study of VDC embedding problem is under the assumption that switch resource is the bottleneck of data center networks(DCNs).To this end,we not only propose relative cost to evaluate embedding strategy,decouple embedding problem into VM placement with marginal resource assignment and virtual link mapping with decided source-destination based on the property of fat-tree,but also design the traffic aware embedding algorithm(TAE) and first fit virtual link mapping(FFLM) to map virtual data center requests to a physical data center.Simulation results show that TAE+FFLM could increase acceptance rate and reduce network cost(about 49%in the case) at the same time.The traffic aware embedding algorithm reduces the load of core-link traffic and brings the optimization opportunity for data center network energy conservation.     

基于智能网络的租户服务质量保证技术

在多租户虚拟网络环境中,用户对于网络服务的多样性以及性能的稳定性需求并不会随着网络架构和运营模式的升级而削弱,用户需求之间的差异性和动态性对于不同切片间资源的分配和调度效率提出了新的挑战.针对多租户虚拟网络的特殊环境,首先提出了QVR(QoS-Virtual Routing)流量调度算法,同时将用户流量调度与网络虚拟资...     

VNE-AFS：基于人工鱼群的网络虚拟化映射算法

虚拟网络资源映射是云计算研究领域的一个难点问题。以降低底层网络映射开销为目标,提出一种基于人工鱼群的网络虚拟化映射算法VNE-AFS。根据虚拟网络请求对底层网络节点和链路的约束关系建立二进制组合优化模型,并利用人工鱼群算法实现虚拟网络资源向底层网络资源的近似最优映射。实验结果表明,与现有的虚拟网络映射算法相比,该算法有效地降低了底层网络的开销和求解时间,提高了虚拟网络映射的成功率、平均收益和资源利用率。     

TRACTOR: Traffic-aware and power-efficient virtual machine placement in edge-cloud data centers using artificial bee colony optimization

Technology providers heavily exploit the usage of edge-cloud data centers (ECDCs) to meet user demand while the ECDCs are large energy consumers. Concerning the decrease of the energy expenditure of ECDCs, task placement is one of the most prominent solutions for effective allocation and consolidation of such tasks onto physical machine (PM). Such allocation must also consider additional optimizations beyond power and must include other objectives, including network-traffic effectiveness. In this study, we present a multi-objective virtual machine (VM) placement scheme (considering VMs as fog tasks) for ECDCs called TRACTOR , which utilizes an artificial bee colony optimization algorithm for power and network-aware assignment of VMs onto PMs. The proposed scheme aims to minimize the network traffic of the interacting VMs and the power dissipation of the data center's switches and PMs. To evaluate the proposed VM placement solution, the Virtual Layer 2 (VL2) and three-tier network topologies are modeled and integrated into the CloudSim toolkit to justify the effectiveness of the proposed solution in mitigating the network traffic and power consumption of the ECDC. Results indicate that our proposed method is able to reduce power energy consumption by 3.5% while decreasing network traffic and power by 15% and 30%, respectively, without affecting other QoS parameters.     

Multi-service differentiated traffic management optimization strategy in cloud data center

In order to cope with the traffic management for multi-service differentiated in cloud data centers,improving network performance and service experience,the multi-service differentiated (MSD) traffic management model was designed that can suit operational requirements in cloud data center.Fibonacci tree optimization (FTO) algorithm was improved according to the MSD model.MSD-FTO traffic management strategy was proposed in SDN cloud data center.Simulation results show that the strategy takes advantage of FTO global optimization ability and multi-modal adaptive performance.Through the global local alternating optimization of the algorithm,differentiation traffic management schemes are obtained as needed,the problem of multi-services differentiated traffic management is solved in operator cloud data center that improve network performance and service experience in cloud data center effectively.     

Optimized joint cell planning and BS on/off switching for LTE networks

Energy is becoming a main concern nowadays due to the increasing demands on natural energy resources. Base stations (BSs) consume up to 80% of the total energy expenditure in a cellular network. In this paper, we propose and evaluate a green radio network planning approach by jointly optimizing the number of active BSs and the BS on/off switching patterns based on the changing traffic conditions in the network in an effort to reduce the total energy consumption of the BSs. The problem is formulated as an integer optimization problem, which proves to be NP‐complete, and thus it can be efficiently solved for small to medium network sizes. For large network sizes, we propose a heuristic solution with close to optimal performance because the optimal solution becomes computationally complex. Planning is performed based on two approaches: a reactive and a proactive approach. In the proactive approach, planning will be performed starting with the lowest traffic demand until reaching the highest traffic demand, whereas in the reactive approach, the reverse way is considered. Performance results are presented for various case studies and are complemented by testing the proposed approaches using commercial radio network planning tools. Results demonstrate considerable energy savings reaching up to 40% through dynamic adaptation of the number of simultaneously active BSs. Copyright © 2015 John Wiley & Sons, Ltd.     

异构云DCN中绿色虚拟算法设计

为进一步提升异构云数据中心网络(DCN)动态管理的科学性,在总结当前主流研究局限性的基础上构思一种基于全局相对最优化的绿色虚拟算法.算法综合考虑虚拟机迁移过程中可能涉及到的诸多客观因素,通过科学地规划时间门限、主机筛选策略、以及精度比较机制对虚拟机实施高效的迁移.数据考察表明,所部署的算法不仅可快速精确地物色到最适宜的...     

IaaS云中面向能源效率优化的虚拟机调度算法(英文)

In IaaS Cloud,different mapping relationships between virtual machines(VMs) and physical machines(PMs) cause different resource utilization,so how to place VMs on PMs to reduce energy consumption is becoming one of the major concerns for cloud providers.The existing VM scheduling schemes propose optimize PMs or network resources utilization,but few of them attempt to improve the energy efficiency of these two kinds of resources simultaneously.This paper proposes a VM scheduling scheme meeting multiple resource constraints,such as the physical server size(CPU,memory,storage,bandwidth,etc.) and network link capacity to reduce both the numbers of active PMs and network elements so as to finally reduce energy consumption.Since VM scheduling problem is abstracted as a combination of bin packing problem and quadratic assignment problem,which is also known as a classic combinatorial optimization and NP-hard problem.Accordingly,we design a twostage heuristic algorithm to solve the issue,and the simulations show that our solution outperforms the existing PM- or network-only optimization solutions.     

Light trees: optical multicasting for improved performance inwavelength routed networks

We introduce the concept of a light-tree in a wavelength-routed optical network. A light-tree is a point-to-multipoint generalization of a lightpath. A lightpath is a point-to-point all-optical wavelength channel connecting a transmitter at a source node to a receiver at a destination node. Lightpath communication can significantly reduce the number of hops (or lightpaths) a packet has to traverse; and this reduction can, in turn, significantly improve the network's throughput. We extend the lightpath concept by incorporating an optical multicasting capability at the routing nodes in order to increase the logical connectivity of the network and further decrease its hop distance. We refer to such a point-to-multipoint extension as a light-tree. Light-trees can not only provide improved performance for unicast traffic, but they naturally can better support multicast traffic and broadcast traffic. In this study, we shall concentrate on the application and advantages of light-trees to unicast and broadcast traffic. We formulate the light-tree-based virtual topology design problem as an optimization problem with one of two possible objective functions: for a given traffic matrix, (i) minimize the network-wide average packet hop distance, or (ii) minimize the total number of transceivers in the network. We demonstrate that an optimum light-tree-based virtual topology has clear advantages over an optimum lightpath-based virtual topology with respect to the above two objectives     

Programmable virtual network as a service: towards a future Internet testbed based on IaaS cloud

During recently years, several OpenFlow-enabled testbeds have been deployed in world-wide research community. Typically, these OpenFlow-enabled testbeds need to stitch to other testbeds to link to virtual servers and immit experimentation traffic. From the view point of researchers, these OpenFlow-enabled testbeds only provide OpenFlow networking resource, and they have to resort to other experiment infrastructure to provide computing and storage resource. Due to the OpenFlow networking and other resource belongs to different infrastructure provider and is managed by their own control software, it is difficult to coordinate these partners to provide a full programmable experiment environment. Meanwhile, the control software of these testbeds are tight coupling with their substrate resources, which means that these substrate resources, together with virtualization technologies, are permanently dedicated to the control software and difficult to be used by other services. In this paper, a new future Internet testbed architecture based on the open Infrastructure-as-a-service cloud and software defined network (SDN) paradigm is proposed. It extends the current virtual network service by adding programmable virtual switch and controller resources that can be controlled by the researcher. Its loose coupling model allows the testbed operator decouple the experiment service from the infrastructure provider, which is a more flexible way to build the testbed. The initial prototype implementation in this paper shows that this new testbed architecture built on IaaS cloud and SDN is feasible and flexible to provide programmable virtual network service.     

A hybrid virtual network mapping algorithm based on threshold load

The virtual resource management architecture for satellite networks currently suffers from a very poor virtual network mapping success rate. This arises because of the need to map multiple heterogeneous virtual networks to the underlying satellite network. Most heuristic algorithms divide virtual network mapping into node mapping and link mapping, which aims to reduce the complexity of the problem. However, this approach is not well suited to highly dynamic satellite networks. In this paper, we propose a hybrid virtual network mapping algorithm that is based on threshold load. This takes the overall load for the nodes as its optimization objective, and combines the idea of backtracking contained in 1‐stage mapping methods and the idea of global optimization contained in 2‐stage mapping methods. The algorithm reduces the complexity of backtracking computation, while avoiding any incompleteness that might result from separating nodes and link mapping. The success rate for virtual network mapping is thus improved, as is the utilization rate for satellite network resources.     

Network planning based on virtual path bandwidth management

Periodical performance evaluation and adaptive resource assignment, already proposed as performance-oriented management, seems to be the most suitable strategy for network planning under demand uncertainty. In this paper, we exploit the inherent capability of ATM networks to rearrange dynamically the already installed resources, and propose performance-oriented management combined with virtual path bandwidth (VPB) control for the planning of the extensions of bandwidth capacities of virtual paths (VPs) and transmission links of the network. We define a large network optimization problem and solve it by a rigorous, analytical procedure. The optimization model comprises specific requirements of the network-planning problem and a bandwidth distribution scheme ensuring network reliability. We reveal the efficiency of the proposed scheme by applying it on a model network, considering two realistic case-studies of network-traffic evolution. We show that in the presence of VPB control: (a) the initial distribution of the total bandwidth to VPs is of no importance, since it can be adaptively rearranged according to the offered traffic, (b) the network is well used and bandwidth investment could be saved, and (c) whenever additional bandwidth must be installed in VPs which have an unanticipated bad grade-of-service, time savings result. We present the network performance in detail, in figures, and compare this with the performance of the network in the absence of VPB control.     

V2X offloading and resource allocation under SDN and MEC architecture

To address the serious problem of delay and energy consumption increase and service quality degradation caused by complex network status and huge amounts of computing data in the scenario of vehicle-to-everything (V2X),a vehicular network architecture combining mobile edge computing (MEC) and software defined network (SDN) was constructed.MEC sinks cloud serviced to the edge of the wireless network to compensate for the delay fluctuation caused by remote cloud computing.The SDN controller could sense network information from a global perspective,flexibly schedule resources,and control offload traffic.To further reduce the system overhead,a joint task offloading and resource allocation scheme was proposed.By modeling the MEC-based V2X offloading and resource allocation,the optimal offloading decision,communication and computing resource allocation scheme were derived.Considering the NP-hard attribute of the problem,Agglomerative Clustering was used to select the initial offloading node,and Q-learning was used for resource allocation.The offloading decision was modeled as an exact potential game,and the existence of Nash equilibrium was proved by the potential function structure.The simulation results show that,as compared to other mechanisms,the proposed mechanism can effectively reduce the system overhead.     

基于机器学习的无线网络流量预测与增长潜力评估

提出一个基于机器学习的无线网络流量预测及流量增长潜力评估方案。该方案分析蜂窝网络中的实际业务流量数据在时间维度上的变化规律,并借助高斯过程的机器学习方法来预测业务变化趋势,从短期角度为运营商的网络优化部署提供指导。基于极限梯度提升(XGBoost)机器学习框架,建立网络中其他运营数据与业务流量的多维映射关系,应用改进的量子粒子群算法进一步寻找蜂窝小区所能承载的流量上限,从长期角度为网络优化部署提供指导,提升网络流量水平、释放流量增长潜力。     

基于卷积神经网络的Android流量分类方法

深度学习就是机器学习研究的过程,主要通过模拟人脑分析学习的过程对数据进行分析。目前,深度学习技术已经在计算机视觉、语音识别、自然语言处理等领域获得了较大发展,并且随着该技术的不断发展,为网络流量分类和异常检测带来了新的发展方向。移动智能手机与大家的生活息息相关,但是其存在的安全问题也日益凸显。针对传统机器学习算法对于流量分类需要人工提取特征、计算量大的问题,提出了基于卷积神经网络模型的应用程序流量分类算法。首先,将网络流量数据集进行数据预处理,去除无关数据字段,并使数据满足卷积神经网络的输入特性。其次,设计了一种新的卷积神经网络模型,从网络结构、超参数空间以及参数优化方面入手,构造了最优分类模型。该模型通过卷积层自主学习数据特征,解决了传统基于机器学习的流量分类算法中的特征选择问题。最后,通过CICAndmal2017网络公开数据集进行模型测试,相比于传统的机器学习流量分类模型,设计的卷积神经网络模型的查准率和查全率分别提高了2.93%和11.87%,同时在类精度、召回率以及F1分数方面都有较好的提升。     

A Multi-objective Optimal Task Scheduling in Cloud Environment Using Cuckoo Particle Swarm Optimization

In cloud computing, varied demands are placed on the constantly changing resources. The task scheduling place very vital role in cloud computing environments, this scheduling process needs to schedule the tasks to virtual machine while reducing the makespan and cost. The task scheduling problem comes under NP hard category. Efficient scheduling method makes cloud computing services better and faster. In general, optimization algorithms are used to solve the scheduling issues in cloud. So, in this paper we combined two optimization algorithms namely called as Cuckoo Search (CS) and Particle Swarm Optimization (PSO).The new proposed hybrid algorithm is called as, CS and particle swarm optimization (CPSO). Our main purpose of the proposed paper is to reduce the makespan, cost and deadline violation rate. The performance of the proposed CPSO algorithm is evaluated using cloudsim toolkit. From the simulation results our proposed works minimize the makespan, cost, deadline violation rate, when compared to PBACO, ACO, MIN–MIN, and FCFS.

     

云数据中心的安全虚拟网络

虚拟网络安全是云计算安全的重要组成。为了保障虚拟网络流量的可控性和安全性,文中提出了一种基于Ethsec加密压缩技术的安全虚拟网络解决方案。该方案设计了虚拟化安全层、虚拟化安全交换机、安全虚拟网络管理平台和安全虚拟网络密钥分发系统等组件,通过文中提出的Ethsec技术,采用国产商用密码算法SM2和SM4算法,对虚拟机的以太网MAC帧进行压缩和解密,实现虚拟化安全交换机对所有虚拟网络流量的监控和分析。     

边缘网络中一种VNF需求预测方法

针对当前虚拟网络功能（Virtualization Network Functions,VNF）需求预测方法准确率较低且不适用于边缘网络的问题,提出了一种在边缘网络中基于支持向量回归（Support Vector Regression,SVR）与门控循环单元（Gated Recurrent Unit,GRU）神经网络模型结合的VNF需求预测方法。考虑到网络边缘流量具有突发性、自相似性及长相关性等特点,结合SVR和GRU两种模型的优点,利用计算复杂度较低的SVR和GRU模型分别提取网络服务历史时序数据的短期特征和长期特征,以提高VNF需求预测准确率,实现边缘网络中VNF的提前放置。实验表明,所提出的预测方法在边缘网络中针对不同网络服务的预测较于传统方法、循环神经网络（Recurrent Neural Networks,RNN）、长短期记忆网络（Long Short-Term Memory,LSTM）模型能够降低20%~30%的误差,有更佳的预测效果。