您现在访问的是微软AZURE全球版技术文档网站,若需要访问由世纪互联运营的MICROSOFT AZURE中国区技术文档网站,请访问 https://docs.azure.cn.

高性能计算虚拟机大小High performance compute virtual machine sizes

A8-A11 和 H 系列大小也称为“计算密集型实例”。The A8-A11 and H-series sizes are also known as compute-intensive instances. 运行这些大小的硬件专为计算密集型和网络密集型应用程序而设计和优化,包括高性能计算 (HPC) 群集应用程序、建模和模拟。The hardware that runs these sizes is designed and optimized for compute-intensive and network-intensive applications, including high-performance computing (HPC) cluster applications, modeling, and simulations. A8-A11 系列使用 Intel Xeon E5-2670 @ 2.6 GHZ,H 系列使用 Intel Xeon E5-2667 v3 @ 3.2 GHz。The A8-A11 series uses Intel Xeon E5-2670 @ 2.6 GHZ and the H-series uses Intel Xeon E5-2667 v3 @ 3.2 GHz. 本文针对此分组中每种大小提供有关 vCPU、数据磁盘和 NIC 的数量,以及存储吞吐量和网络带宽的信息。This article provides information about the number of vCPUs, data disks, and NICs as well as storage throughput and network bandwidth for each size in this grouping.

Azure H 系列虚拟机是最新高性能计算 VM,旨在满足高端计算需求,例如分子建模和计算流体动力学。Azure H-series virtual machines are the latest in high performance computing VMs aimed at high end computational needs, like molecular modeling, and computational fluid dynamics. 这些 8 vCPU 和 16 vCPU 的 VM 基于采用 DDR4 内存和基于 SSD 的临时存储的 Intel Haswell E5-2667 V3 处理器技术构建。These 8 and 16 vCPU VMs are built on the Intel Haswell E5-2667 V3 processor technology featuring DDR4 memory and SSD-based temporary storage.

除强大的 CPU 功能外,H 系列还提供支持使用 InfiniBand 的低延迟 RDMA 网络的各种选项,以及支持内存密集型计算要求的多项内存配置。In addition to the substantial CPU power, the H-series offers diverse options for low latency RDMA networking using FDR InfiniBand and several memory configurations to support memory intensive computational requirements.

H 系列H-series

ACU:290-300ACU: 290-300

大小Size vCPUvCPU 内存:GiBMemory: GiB 临时存储 (SSD) GiBTemp storage (SSD) GiB 最大数据磁盘数Max data disks 磁盘最大吞吐量:IOPSMax disk throughput: IOPS 最大 NIC 数Max NICs
Standard_H8Standard_H8 88 5656 10001000 3232 32 x 50032 x 500 22
Standard_H16Standard_H16 1616 112112 20002000 6464 64 x 50064 x 500 44
Standard_H8mStandard_H8m 88 112112 10001000 3232 32 x 50032 x 500 22
Standard_H16mStandard_H16m 1616 224224 20002000 6464 64 x 50064 x 500 44
Standard_H16r 1Standard_H16r 1 1616 112112 20002000 6464 64 x 50064 x 500 44
Standard_H16mr 1Standard_H16mr 1 1616 224224 20002000 6464 64 x 50064 x 500 44

1 对于 MPI 应用程序来说,专用 RDMA 后端网络是通过 FDR InfiniBand 网络启用的,后者可以提供相当低的延迟和高带宽。1 For MPI applications, dedicated RDMA backend network is enabled by FDR InfiniBand network, which delivers ultra-low-latency and high bandwidth.


A 系列 - 计算密集型实例A-series - compute-intensive instances

ACU:225ACU: 225

大小Size vCPUvCPU 内存:GiBMemory: GiB 临时存储 (HDD):GiBTemp storage (HDD): GiB 最大数据磁盘数Max data disks 数据磁盘最大吞吐量:IOPSMax data disk throughput: IOPS 最大 NIC 数Max NICs
Standard_A8 1Standard_A8 1 88 5656 382382 3232 32x50032x500 22
Standard_A9 1Standard_A9 1 1616 112112 382382 6464 64x50064x500 44
Standard_A10Standard_A10 88 5656 382382 3232 32x50032x500 22
Standard_A11Standard_A11 1616 112112 382382 6464 64x50064x500 44

1对于 MPI 应用程序来说,专用 RDMA 后端网络是通过 FDR InfiniBand 网络启用的,后者可以提供相当低的延迟和高带宽。1For MPI applications, dedicated RDMA backend network is enabled by FDR InfiniBand network, which delivers ultra-low-latency and high bandwidth.


大小表定义Size table definitions

  • 存储容量的单位为 GiB 或 1024^3 字节。Storage capacity is shown in units of GiB or 1024^3 bytes. 比较以 GB(1000^3 字节)为单位的磁盘和以 GiB(1024^3 字节)为单位的磁盘时,请记住以 GiB 为单位的容量数显得更小。When comparing disks measured in GB (1000^3 bytes) to disks measured in GiB (1024^3) remember that capacity numbers given in GiB may appear smaller. 例如,1023 GiB = 1098.4 GBFor example, 1023 GiB = 1098.4 GB
  • 磁盘吞吐量的单位为每秒输入/输出操作数 (IOPS) 和 MBps,其中 MBps = 10^6 字节/秒。Disk throughput is measured in input/output operations per second (IOPS) and MBps where MBps = 10^6 bytes/sec.
  • 数据磁盘可以在缓存或非缓存模式下运行。Data disks can operate in cached or uncached modes. 对于缓存数据磁盘操作,主机缓存模式设置为 ReadOnlyReadWriteFor cached data disk operation, the host cache mode is set to ReadOnly or ReadWrite. 对于非缓存数据磁盘操作,主机缓存模式设置为 NoneFor uncached data disk operation, the host cache mode is set to None.
  • 若要获得 VM 的最佳性能,应将数据磁盘数限制为每 vCPU 2 个磁盘。If you want to get the best performance for your VMs, you should limit the number of data disks to 2 disks per vCPU.
  • 预期的网络带宽是指跨所有 NIC 为每个 VM 类型分配的最大聚合带宽,适用于所有目标。Expected network bandwidth is the maximum aggregated bandwidth allocated per VM type across all NICs, for all destinations. 不保证能够达到上述预期上限,但目的是提供相关指导,以便为目标应用程序选择适当的 VM 类型。Upper limits are not guaranteed, but are intended to provide guidance for selecting the right VM type for the intended application. 实际的网络性能将取决于许多因素,包括网络拥塞、应用程序负载和网络设置。Actual network performance will depend on a variety of factors including network congestion, application loads, and network settings. 有关如何优化网络吞吐量的信息,请参阅为 Windows 和 Linux 优化网络吞吐量For information on optimizing network throughput, see Optimizing network throughput for Windows and Linux. 若要在 Linux 或 Windows 中达到预期的网络性能,可能需要选择特定的版本,或者需要优化 VM。To achieve the expected network performance on Linux or Windows, it may be necessary to select a specific version or optimize your VM. 有关详细信息,请参阅如何针对虚拟机吞吐量进行可靠的测试For more information, see How to reliably test for virtual machine throughput.

部署注意事项Deployment considerations

  • Azure 订阅 - 若要部署不止一些计算密集型实例,请考虑使用即用即付订阅或其他购买选项。Azure subscription – To deploy more than a few compute-intensive instances, consider a pay-as-you-go subscription or other purchase options. 如果使用的是 Azure 免费帐户,则仅可以使用有限数量的 Azure 计算核心。If you're using an Azure free account, you can use only a limited number of Azure compute cores.

  • 定价和可用性 - 只在标准定价层提供 VM 大小。Pricing and availability - These VM sizes are offered only in the Standard pricing tier. 有关各 Azure 区域推出的产品,请查看 按区域提供的产品Check Products available by region for availability in Azure regions.

  • 核心配额 - 可能需要在 Azure 订阅中在默认值的基础上增加核心配额。Cores quota – You might need to increase the cores quota in your Azure subscription from the default value. 订阅可能也会限制可在特定 VM 大小系列(包括 H 系列)中部署的核心数目。Your subscription might also limit the number of cores you can deploy in certain VM size families, including the H-series. 若要请求提高配额,可免费提出在线客户支持请求To request a quota increase, open an online customer support request at no charge. (默认限制可能会因订阅类别而异。)(Default limits may vary depending on your subscription category.)

    备注

    如果有大规模容量需求,请联系 Azure 支持。Contact Azure Support if you have large-scale capacity needs. Azure 配额为信用额度,而不是容量保障。Azure quotas are credit limits, not capacity guarantees. 不管配额是什么,都只根据所用的核心数计费。Regardless of your quota, you are only charged for cores that you use.

  • 虚拟网络 – Azure 虚拟网络不需要使用计算密集型实例。Virtual network – An Azure virtual network is not required to use the compute-intensive instances. 但是,对于许多部署来说,如果需要访问本地资源,则可能至少需要一个基于云的 Azure 虚拟网络或站点到站点连接。However, for many deployments you need at least a cloud-based Azure virtual network, or a site-to-site connection if you need to access on-premises resources. 需要时,请创建一个新的虚拟网络来部署实例。When needed, create a new virtual network to deploy the instances. 不支持将计算密集型 VM 添加到地缘组中的虚拟网络。Adding compute-intensive VMs to a virtual network in an affinity group is not supported.
  • 调整大小 - 考虑到专用硬件,可以只对同一大小系列(H 系列或计算密集型 A 系列)内的计算密集型实例进行大小调整。Resizing – Because of their specialized hardware, you can only resize compute-intensive instances within the same size family (H-series or compute-intensive A-series). 例如,可仅将 H 系列 VM 的大小从一个 H 系列大小调整为另一个。For example, you can only resize an H-series VM from one H-series size to another. 此外,不支持从非计算密集型大小调整为计算密集型大小。In addition, resizing from a non-compute-intensive size to a compute-intensive size is not supported.

支持 RDMA 的实例RDMA-capable instances

计算密集型实例(H16r、H16mr、A8 和 A9)的子集具有远程直接内存访问 (RDMA) 连接的网络接口。A subset of the compute-intensive instances (H16r, H16mr, A8, and A9) feature a network interface for remote direct memory access (RDMA) connectivity. (使用“r”指定的所选 N 系列大小(如 NC24r)也支持 RDMA 功能。)此接口是对可供其他 VM 大小使用的标准 Azure 网络接口的补充。(Selected N-series sizes designated with 'r' such as NC24r are also RDMA-capable.) This interface is in addition to the standard Azure network interface available to other VM sizes.

此接口允许支持 RDMA 的实例通过 InfiniBand (IB) 网络进行通信,对 H16r、H16mr 和支持 RDMA 的 N 系列虚拟机以 FDR 速率运行,对 A8 和 A9 虚拟机以 QDR 速率运行。This interface allows the RDMA-capable instances to communicate over an InfiniBand (IB) network, operating at FDR rates for H16r, H16mr, and RDMA-capable N-series virtual machines, and QDR rates for A8 and A9 virtual machines. 这些 RDMA 功能可以提高某些消息传递接口 (MPI) 应用程序的可伸缩性和性能。These RDMA capabilities can boost the scalability and performance of certain Message Passing Interface (MPI) applications.

备注

在 Azure 中,不支持基于 IB 的 IP。In Azure, IP over IB is not supported. 仅支持 RDMA over IB。Only RDMA over IB is supported.

在同一个可用性集或 VM 规模集中(使用 Azure 资源管理器部署模型时)或同一个云服务中(使用经典部署模型时)部署支持 RDMA 的 HPC VM。Deploy the RDMA-capable HPC VMs in the same availability set or VM scale set (when you use the Azure Resource Manager deployment model) or the same cloud service (when you use the classic deployment model). 支持 RDMA 的 HPC VM 可以访问 Azure RDMA 网络的其他要求如下。Additional requirements for RDMA-capable HPC VMs to access the Azure RDMA network follow.

MPIMPI

仅支持 Intel MPI 5.x 版本。Only Intel MPI 5.x versions are supported. 更高版本(2017 版和 2018 版)的 Intel MPI 运行时库与 Azure Linux RDMA 驱动程序不兼容。Later versions (2017, 2018) of the Intel MPI runtime library are not compatible with the Azure Linux RDMA drivers.

分发Distributions

从支持 RDMA 连接的 Azure Marketplace 中的一个映像部署计算密集型 VM:Deploy a compute-intensive VM from one of the images in the Azure Marketplace that supports RDMA connectivity:

  • Ubuntu - Ubuntu Server 16.04 LTS。Ubuntu - Ubuntu Server 16.04 LTS. 在 VM 上配置 RDMA 驱动程序,并注册 Intel 下载 Intel MPI:Configure RDMA drivers on the VM and register with Intel to download Intel MPI:

    1. 安装 dapl、rdmacm、ibverbs 和 mlx4Install dapl, rdmacm, ibverbs, and mlx4

      sudo apt-get update
      
      sudo apt-get install libdapl2 libmlx4-1
      
    2. 在 /etc/waagent.conf 中,通过取消注释以下配置行来启用 RDMA。In /etc/waagent.conf, enable RDMA by uncommenting the following configuration lines. 需要根访问权限才能编辑该文件。You need root access to edit this file.

      OS.EnableRDMA=y
      
      OS.UpdateRdmaDriver=y
      
    3. 在 /etc/security/limits.conf 文件中,添加或更改 KB 中的以下内存设置。Add or change the following memory settings in KB in the /etc/security/limits.conf file. 需要根访问权限才能编辑该文件。You need root access to edit this file. 出于测试目的,可以将 memlock 设置为不受限制。For testing purposes you can set memlock to unlimited. 例如:<User or group name> hard memlock unlimitedFor example: <User or group name> hard memlock unlimited.

      <User or group name> hard    memlock <memory required for your application in KB>
      
      <User or group name> soft    memlock <memory required for your application in KB>
      
    4. 安装 Intel MPI 库。Install Intel MPI Library. 从 Intel 购买和下载库或下载免费评估版本Either purchase and download the library from Intel or download the free evaluation version.

      wget http://registrationcenter-download.intel.com/akdlm/irc_nas/tec/11595/l_mpi_2017.3.196.tgz
      

      有关安装步骤,请参阅 Intel MPI 库安装指南For installation steps, see the Intel MPI Library Installation Guide.

    5. 启用非根非调试器进程的 ptrace(为最新版本的 Intel MPI 所需)。Enable ptrace for non-root non-debugger processes (needed for the most recent versions of Intel MPI).

      echo 0 | sudo tee /proc/sys/kernel/yama/ptrace_scope
      
  • SUSE Linux Enterprise Server - SLES 12 SP3 for HPC、SLES 12 SP3 for HPC(高级版)、SLES 12 SP1 for HPC、SLES 12 SP1 for HPC(高级版)。SUSE Linux Enterprise Server - SLES 12 SP3 for HPC, SLES 12 SP3 for HPC (Premium), SLES 12 SP1 for HPC, SLES 12 SP1 for HPC (Premium). 安装 RDMA 驱动程序,并在 VM 上分发 Intel MPI 包。RDMA drivers are installed and Intel MPI packages are distributed on the VM. 运行以下命令安装 MPI:Install MPI by running the following command:

    sudo rpm -v -i --nodeps /opt/intelMPI/intel_mpi_packages/*.rpm
    
  • 基于 CentOS 的 HPC - 基于 CentOS 的 6.5 HPC 或更高版本(对于 H 系列,建议使用版本 7.1 或更高版本)。CentOS-based HPC - CentOS-based 6.5 HPC or a later version (for H-series, version 7.1 or later is recommended). 在 VM 上安装 RDMA 驱动程序和 Intel MPI 5.1。RDMA drivers and Intel MPI 5.1 are installed on the VM.

    备注

    在基于 CentOS 的 HPC 映像中,内核更新已在 yum 配置文件中禁用。On the CentOS-based HPC images, kernel updates are disabled in the yum configuration file. 这是因为 Linux RDMA 驱动程序以 RPM 包的形式分发,如果更新了内核,驱动程序更新可能无法工作。This is because the Linux RDMA drivers are distributed as an RPM package, and driver updates might not work if the kernel is updated.

群集配置Cluster configuration

需要进行额外的系统配置才能在群集 VM 上运行 MPI 作业。Additional system configuration is needed to run MPI jobs on clustered VMs. 例如,在 VM 群集上,需要在计算节点之间建立信任。For example, on a cluster of VMs, you need to establish trust among the compute nodes. 有关典型设置,请参阅 Set up a Linux RDMA cluster to run MPI applications(设置 Linux RDMA 群集以运行 MPI 应用程序)。For typical settings, see Set up a Linux RDMA cluster to run MPI applications.

网络拓扑注意事项Network topology considerations

  • 在 Azure 中支持 RDMA 的 Linux VM 上,Eth1 保留用于传输 RDMA 网络流量。On RDMA-enabled Linux VMs in Azure, Eth1 is reserved for RDMA network traffic. 不要更改引用此网络的配置文件中的任何 Eth1 设置或任何信息。Do not change any Eth1 settings or any information in the configuration file referring to this network. Eth0 保留用于传输常规 Azure 网络流量。Eth0 is reserved for regular Azure network traffic.

  • Azure 中的 RDMA 网络保留地址空间 172.16.0.0/16。The RDMA network in Azure reserves the address space 172.16.0.0/16.

使用 HPC PackUsing HPC Pack

HPC Pack 是 Microsoft 的免费 HPC 群集和作业管理解决方案,可以通过它在 Linux 上使用计算密集型实例。HPC Pack, Microsoft’s free HPC cluster and job management solution, is one option for you to use the compute-intensive instances with Linux. 最新版本的 HPC Pack 支持多个 Linux 分发在由 Windows Server 头节点管理的 Azure VM 中部署的计算节点上运行。The latest releases of HPC Pack support several Linux distributions to run on compute nodes deployed in Azure VMs, managed by a Windows Server head node. 使用支持 RDMA 且运行 Intel MPI 的 Linux 计算节点,HPC Pack 可以计划和运行访问 RDMA 网络的 Linux MPI 应用程序。With RDMA-capable Linux compute nodes running Intel MPI, HPC Pack can schedule and run Linux MPI applications that access the RDMA network. 请参阅 Get started with Linux compute nodes in an HPC Pack cluster in Azure(Azure 的 HPC Pack 群集中的 Linux 计算节点入门)。See Get started with Linux compute nodes in an HPC Pack cluster in Azure.

其他大小Other sizes

后续步骤Next steps