基于跨虚拟机零下陷通信的加速器虚拟化框架

人工智能技术的长足发展对于云计算的算力提出了更高的要求,云服务提供商在数据中心内添置了拥有大量并行计算单元的加速器,这些加速器需要与已有的虚拟化平台相结合以进行计算资源的划分.当前主流的加速器虚拟化方案是通过PCI透传的方式,但是该方式不支持细粒度的资源划分;部分特定型号的加速器还支持了时分复用的方案,通过硬件与虚拟机监视器配合划分计算资源和时间片,但是该方案可移植性差,对于任何新型加速器的适配都要重新开发,固定的资源划分策略也导致可扩展性有限;另有基于API转发的方案,通过分离式驱动的模式将虚拟机的请求转发给后端驱动处理,而转发通信的过程中存在着性能瓶颈.提出了Wormhole,一种基于C/S架构的、支持跨虚拟机快速代理执行的加速器虚拟化框架,旨在为上层用户提供高效、透明的加速器API转发虚拟化的同时保障多用户间的强隔离性.该框架利用硬件虚拟化技术,允许CPU控制流在虚拟机间快速切换而不触发任何下陷,大幅降低了虚拟机间通信带来的虚拟化性能开销.实验结果表明,Wormhole的原型系统相较于具有代表性的开源虚拟化方案GvirtuS,在经典模型的训练测试中能够有高达5倍的性能提升.     

一种灵活高效的虚拟CPU调度算法

目前,虚拟化已经广泛应用于数据中心,但主流的虚拟CPU调度策略并没有实现对I/O性能的保障,尤其当延时敏感型负载的虚拟机和计算敏感型负载的虚拟机竞争CPU资源时,其性能显著下降.针对上述问题,本文提出了一种灵活、高效的虚拟CPU调度算法（FLMS）.FLMS通过采用虚拟机分类、虚拟CPU绑定、多类时间片等技术降低了虚拟机的响应延时,同时基于多处理器架构重新设计了负载均衡策略,优化了虚拟CPU迁移.FLMS通用于目前主流的虚拟化方案,在软件虚拟化方式下相比于最新的优化方案延时降低了30%,带宽有10%的提升;在使用硬件辅助虚拟化的系统中,通过FLMS能够获得接近原生系统的I/O性能,并且保证了整个系统的公平性.     

自适应调整虚拟机权重参数的调度方法

在基于特权服务操作系统的虚拟机架构下客户操作系统需要借助特权服务操作系统来访问真实硬件,目前虚拟机调度算法的优化主要是侧重于I/O密集型虚拟机的研究,而忽视了CPU密集型虚拟机,更忽视了特权服务操作系统的I/O处理能力对虚拟机整体性能的影响.针对这些问题,提出了一种基于Credit算法的自适应调整虚拟机权重参数的优化调度方法,将特权服务操作系统的I/O处理能力作为虚拟机参数调整的一个重要参数,同时兼顾I/O密集型虚拟机和CPU密集型虚拟机对资源的需求.实验结果表明该方法能够及时根据当前的I/O请求数量和特权服务操作系统的处理能力合理调整虚拟机的权重参数,从而大大提高了客户操作系统CPU处理性能和硬件设备的访问性能.     

异构云平台中能源有效的虚拟机部署研究

能源消耗已经成为数据中心操作成本的重要组成部分,虚拟化技术是降低数据中心能源消耗的有效方法之一.为了降低数据中心过高的能源消耗,利用虚拟化技术,结合数据中心中物理机的异构性和虚拟机所需资源的多维性,提出了一个衡量不同类型物理机性能的模型和一个衡量多维资源利用率的模型,在此基础上提出了一个异构云平台下能源有效的虚拟机部署算法.仿真实验表明,与MBFD算法及BFD算法相比,该算法不仅可以有效地降低系统的能源消耗,而且还提高了资源利用率,减少了资源的浪费.     

面向应用服务级目标的虚拟化资源管理

在虚拟环境中实现应用服务级目标,是当前数据中心系统管理的关键问题之一.解决该问题有两个方面的要求:一方面,在虚拟化层次和范围内,能够动态和分布式地按需调整虚拟机资源分配;另一方面,在虚拟化范围之外,能够控制由于虚拟机对非虚拟化资源的竞争所导致的性能干扰,实现虚拟机性能隔离.然而,已有工作不适用于虚拟化数据中心场景.提出一种面向应用服务级目标的虚拟化资源管理方法.首先,该方法基于反馈控制理论,通过动态调整虚拟机资源分配来实现每个应用的服务器目标;同时,还设计了一个两层结构的自适应机制,使得应用模型能够动态地捕捉虚拟机资源分配与应用性能的时变非线性关系;最后,该方法通过仲裁不同应用的资源分配请求来控制虚拟机在非虚拟化资源上的竞争干扰.实验在基于Xen的机群环境中检验了该方法在RUBiS系统和TPC-W基准上的效果.实验结果显示,该方法的应用服务级目标实现率比两种对比方法平均高29.2%,而应用服务级目标平均偏离率比它们平均低50.1%.另一方面,当RUBiS系统和TPC-W基准竞争非虚拟化的磁盘I/O资源时,该方法通过抑制TPC-W基准28.7%的处理器资源需求来优先满足RUBiS系统的磁盘I/O需求.     

资源调度等待开销感知的虚拟机整合

近年来,数据中心庞大的能源开销问题引起广泛关注.虚拟化管理平台可以通过虚拟机迁移技术将虚拟机整合到更少的服务器上,从而提高数据中心能源有效性.对面向数据中心节能的虚拟机整合研究工作进行调研,并总结虚拟机整合研究存在的3个挑战.针对已有工作未考虑虚拟机等待资源调度带来的服务器资源额外开销这种现象,开展了资源调度等待开销感知的虚拟机整合研究.从理论和实验上证明了在具有实际意义的约束条件下,存在着虚拟机等待资源调度带来的服务器资源额外开销,且随着整合虚拟机数量的增长保持稳定.基于典型工作负载的实验结果表明,这个额外开销平均占据了11.7%的服务器资源开销.此外,提出了资源预留整合（MRC）算法,用于改进已有的虚拟机整合算法.算法模拟实验结果表明,MRC算法相比于常用的虚拟机整合算法FFD（first fit decreasing）,明显降低了服务器资源溢出概率.     

虚拟网络的管理面临多种挑战

虚拟化的优势一直都很明显。它可以提高服务器、计算及网络容量的使用效率,减少资金投入。但是,尽管虚拟化有明显好处,但是它也给负责管理虚拟网络的数据中心人员带来了新的挑战。此外,此外还有满足和维持最终用户性能和保护重要数据等要求。由于虚拟化支持在一台服务器上同时运行多个虚拟机（VM）,然后每一个虚拟机都能运行一套完整的操作系统环境和应用程序,因此它可以提高服务器CPU的使用率。     

虚拟化环境中内存隔离性测试与分析

设计一个虚拟化内存性能隔离度评测系统,该系统采用细粒度的干扰方法,针对虚拟化系统的内存带宽和内存空间的隔离性进行干扰测试,测试多个虚拟机同时访问内存时性能之间的相互影响。测试结果表明,当前的硬件和软件平台在虚拟化性能隔离度的支持方面依然存在着较大的缺陷,在存在干扰情况下,被测虚拟机的性能指标下降严重。     

浅论服务器虚拟化

0前言将服务器的物理资源,通过虚拟技术变成多台可以相互隔离的虚拟服务器,不受限于物理上的界限,而是让CPU、内存、磁盘、I/O等硬件变成可以动态管理的共享资源。从而提高资源的利用率,简化系统管理,实现服务器整--这就是服务器的虚拟化。1服务器虚拟化的类型虚拟化管理所需的硬件设备为虚拟机监视器,软件是通过虚拟化平台进行管理。虚拟机监视器(Virtual Machine Monitor)把     

基于T2080处理器平台的虚拟机管理器设计

随着嵌入式计算机技术的发展，计算资源越来越丰富并提供硬件辅助虚拟化支持能力。为了提升计算效能，通过虚拟化技术实现在一个嵌入式计算机中集成多种异构系统应用，从而降低系统成本，但会带来安全隔离、不同应用的竞争冲突、数据安全、实时性保障等问题。虚拟机管理器借助硬件辅助虚拟化技术，在软件层构建多个资源深度隔离的虚拟机，由操作系统管理虚拟机内的硬件资源、软件资源及应用。本文在T2080处理器平台上设计了虚拟机管理器，实现了处理器核虚拟化、内存虚拟化、设备虚拟化，提供虚拟机管理和安全隔离能力。     

StopGap: elastic VMs to enhance server consolidation

Virtualized cloud infrastructures (also known as IaaS platforms) generally rely on a server consolidation system to pack virtual machines (VMs) on as few servers as possible. However, an important limitation of consolidation is not addressed by such systems. Because the managed VMs may be of various sizes (small, medium, large, etc.), VM packing may be obstructed when VMs do not fit available spaces. This phenomenon leaves servers with a set of unused resources (‘holes’). It is similar to memory fragmentation, a well‐known problem in operating system domain. In this paper, we propose a solution which consists in resizing VMs so that they can fit with holes. This operation leads to the management of what we call elastic VMs and requires cooperation between the application level and the IaaS level, because it impacts management at both levels. To this end, we propose a new resource negotiation and allocation model in the IaaS, called HRNM. We demonstrate HRNM's applicability through the implementation of a prototype compatible with two main IaaS managers (OpenStack and OpenNebula). By performing thorough experiments with SPECvirt_sc2010 (a reference benchmark for server consolidation), we show that the impact of HRNM on customer's application is negligible. Finally, using Google data center traces, we show an improvement of about 62.5% for the traditional consolidation engines. Copyright © 2017 John Wiley & Sons, Ltd.     

Partial migration technique for GPGPU tasks to Prevent GPU Memory Starvation in RPC-based GPU Virtualization

Graphics processing unit (GPU) virtualization technology enables a single GPU to be shared among multiple virtual machines (VMs), thereby allowing multiple VMs to perform GPU operations simultaneously with a single GPU. Because GPUs exhibit lower resource scalability than central processing units (CPUs), memory, and storage, many VMs encounter resource shortages while running GPU operations concurrently, implying that the VM performing the GPU operation must wait to use the GPU. In this paper, we propose a partial migration technique for general-purpose graphics processing unit (GPGPU) tasks to prevent the GPU resource shortage in a remote procedure call-based GPU virtualization environment. The proposed method allows a GPGPU task to be migrated to another physical server's GPU based on the available resources of the target's GPU device, thereby reducing the wait time of the VM to use the GPU. With this approach, we prevent resource shortages and minimize performance degradation for GPGPU operations running on multiple VMs. Our proposed method can prevent GPU memory shortage, improve GPGPU task performance by up to 14%, and improve GPU computational performance by up to 82%. In addition, experiments show that the migration of GPGPU tasks minimizes the impact on other VMs.     

应用内存超售在虚拟化环境中优化SGX的性能

随着安全性在云计算中越来越受到关注,英特尔自2015年起提出了SGX.它提供enclave,并保护enclave中的应用程序免受不信任的软件(包括客户操作系统和虚拟机监视器)和硬件(英特尔CPU包除外)的攻击.然而,SGX只能支持256MB的enclave内存EPC.因此,在不同的虚拟机之间高效分配宝贵的EPC资源对整...     

Multi-objective Swarm Intelligence schedulers for online scientific Clouds

Cloud Computing is a promising paradigm for parallel computing. However, as Cloud-based services become more dynamic, resource provisioning in Clouds becomes more challenging. The paradigm, with its promise of virtually infinite resources, seems to suit well in solving resource greedy scientific computing problems. In a Cloud, an appropriate number of Virtual Machines (VM) is created and allocated in physical resources for executing jobs. This work focuses on the Infrastructure as a Service (IaaS) model where custom VMs are launched in appropriate hosts available in a Cloud to execute scientific experiments coming from multiple users. Finding optimal solutions to allocate VMs to physical resources is an NP-complete problem, and therefore many heuristics have been developed. In this work, we describe and evaluate two Cloud schedulers based on Swarm Intelligence (SI) techniques, particularly Ant Colony Optimization (ACO) and Particle Swarm Optimization (PSO). The main performance metrics to study are the number of serviced users by the Cloud and the total number of created VMs in online (non-batch) scheduling scenarios. We also perform a sensitivity analysis by varying the specific-parameter values of each algorithm to evaluate the impact on the performance of the two objective metrics. The intra-Cloud network traffic is also measured. Simulated experiments performed using CloudSim and job data from real scientific problems show that the use of SI-based techniques succeeds in balancing the studied metrics compared to Genetic Algorithms.     

A Cost Model for IaaS Clouds Based on Virtual Machine Energy Consumption

Cloud Computing has revolutionized the software, platform and infrastructure provisioning. Infrastructure-as-a-Service (IaaS) providers offer on-demand and configurable Virtual Machine (VMs) to tenants of cloud computing services. A key consolidation force that widespread IaaS deployment is the use of pay-as-you-go and pay-as-you-use cost models. In these models, a service price can be composed of two dimensions: the individual consumption, and a proportional value charged for service maintenance. A common practice for public providers is to dilute both capital and operational costs on predefined pricing sheets. In this context, we propose PSVE (Proportional-Shared Virtual Energy), a cost model for IaaS providers based on CPU energy consumption. Aligned with traditional commodity prices, PSVE is composed of two key elements: an individualized cost accounted from CPU usage of VMs (e.g., processing and networking), and a shared cost from common hypervisor management operations, proportionally distributed among VMs.     

Precise contention-aware performance prediction on virtualized multicore system

Multicore systems are widely deployed in both the embedded and the high end computing infrastructures. However, traditional virtualization systems can not effectively isolate shared micro architectural resources among virtual machines (VMs) running on multicore systems. CPU and memory intensive VMs contending for these resources will lead to serious performance interference, which makes virtualization systems less efficient and VM performance less stable. In this paper, we propose a contention-aware performance prediction model on the virtualized multicore systems to quantify the performance degradation of VMs. First, we identify the performance interference factors and design synthetic micro-benchmarks to obtain VM’s contention sensitivity and intensity features that are correlated with VM performance degradation. Second, based on the contention features, we build VM performance prediction model using machine learning techniques to quantify the precise levels of performance degradation. The proposed model can be used to optimize VM performance on multicore systems. Our experimental results show that the performance prediction model achieves high accuracy and the mean absolute error is 2.83%.     

Utilization of workflow management system for virtual machine instance management on cloud

The concept of virtualization is one of the most important technologies to construct a cloud service, and especially hardware virtualization is indispensable for infrastructure as a service (IaaS) where the cloud offering, infrastructure, is usually provided as a pool of virtual machine (VM) instances. For that reason, many public IaaS clouds like Amazon Web Service and private cloud toolkits such as Eucalyptus and OpenStack provide users with methods for managing VM instances via APIs, command‐line tools, web services, and so on. These are, however, not easy to use or customize for the average end users, especially for those in scientific research areas who just want to perform their work on a cloud and do not need to know the underlying technologies that much. Utilizing workflow management systems (WfMSs) in managing VMs on a cloud can alleviate these difficulties. Users only need to describe parameters needed for VMs and enact the workflow on a workflow enactment engine using user‐friendly interfaces. We propose a management scheme for VM instances on a cloud with the WfMS in this paper. We present a preliminary study on integrating cloud and WfMS focusing on management of VM instances and show an early implementation for a proof of concept with detailed explanations and possible usage scenarios. Copyright © 2015 John Wiley & Sons, Ltd.     

基于改进多目标布谷鸟搜索的资源调度算法

针对IaaS(Infrastructure as a Service)云计算中资源调度的多目标优化问题,提出一种基于改进多目标布谷鸟搜索的资源调度算法。在多目标布谷鸟搜索算法的基础上,通过改进随机游走策略和丢弃概率策略提高了算法的局部搜索能力和收敛速度。以最大限度地减少完成时间和成本为主要目标,将任务分配特定的VM(Virtual Manufacturing)满足云用户对云提供商的资源利用的需求,从而减少延迟,提高资源利用率和服务质量。实验结果表明,该算法可以有效地解决IaaS云计算环境中资源调度的多目标问题,与其他算法相比,具有一定的优势。     

Flexible service consolidation with nested virtualization and library operating systems

In Infrastructure-as-a-service (IaaS) clouds, users can reduce costs by scale-in or scale-down when running services are underutilized. Since these optimizations of instance deployment require at least one minimum instance even for running an underutilized service, cost reduction is limited. For further optimization, multiple services can be consolidated into one instance. However, services have to be stopped temporarily at the consolidation time, and isolation between services becomes weaker after the consolidation. To solve these problems, this paper proposes FlexCapsule, which enables seamless and secure service consolidation in existing IaaS clouds. FlexCapsule runs each service in a lightweight virtual machine (VM) called an app VM, using a library operating system. An app VM runs inside an instance using a technique called nested virtualization. FlexCapsule can optimize instance deployment with negligible downtime by flexibly migrating app VMs. Due to strong isolation provided by app VMs, it can guarantee security between consolidated services. In addition, FlexCapsule provides multiprocess support using app VMs by emulating process fork and process pools. We have implemented FlexCapsule in Xen using both fully virtualized OS^v and paravirtualized MiniOS. Then, we examined the effectiveness of FlexCapsule using several applications. Due to the premature implementation of nested virtualization in Xen, the performance of app VMs largely degraded, but we believe that the performance could be improved using several existing optimizations.