Azure has two different deployment models for creating and working with resources: Resource Manager and classic. This article covers using both models, but Microsoft recommends that most new deployments use the Resource Manager model.
This article provides background information and considerations for using the Azure A8, A9, A10, and A11 instances, also known as compute-intensive instances. Key features of these instances include:
High-performance hardware – The Azure datacenter hardware that runs these instances is designed and optimized for compute-intensive and network-intensive applications, including high-performance computing (HPC) cluster applications, modeling, and simulations.
RDMA network connection for MPI applications – When configured with the necessary network drivers, the A8 and A9 instances can communicate with other A8 and A9 instances over a low-latency, high-throughput network in Azure that is based on remote direct memory access (RDMA) technology. This feature can boost the performance of applications that use supported Linux or Windows Message Passing Interface (MPI) implementations.
Support for Linux and Windows HPC clusters – Deploy cluster management and job scheduling software on the A8, A9, A10, and A11 instances in Azure to create a stand-alone HPC cluster or to add capacity to an on-premises cluster. Like other Azure VM sizes, the A8, A9, A10, and A11 instances support standard or custom Windows Server and Linux operating system images or Azure Resource Manager templates in Azure VMs (IaaS), or Azure Guest OS releases in cloud services (PaaS, for Windows Server only).
A10 and A11 instances have the same performance optimizations and specifications as the A8 and A9 instances. However, they do not include access to the RDMA network in Azure. They are designed for HPC applications that do not require constant and low-latency communication between nodes, also known as parametric or embarrassingly parallel applications.
The Azure A8, A9, A10, and A11 compute-intensive instances feature high-speed, multicore CPUs and large amounts of memory, as shown in the following table.
|A8 and A10||Intel Xeon E5-2670
8 cores @ 2.6 GHz
|A9 and A11||Intel Xeon E5-2670
16 cores @ 2.6 GHz
Additional processor details, including supported instruction set extensions, are at the Intel.com website. For VM storage capacities and disk details, see Sizes for virtual machines.
A8 and A9 instances have two network adapters, which connect to the following two back-end Azure networks.
|10-Gbps Ethernet||Connects to Azure services (such as Azure Storage and Azure Virtual Network) and to the Internet.|
|32-Gbps back end, RDMA capable||Enables low-latency, high-throughput application communication between instances within a single cloud service or availability set. Reserved for MPI traffic only.|
On A8 and A9 VMs running Linux, access to the RDMA network is currently enabled only through applications that use Azure Linux RDMA and Intel MPI Library 5 on SUSE Linux Enterprise Server 12 (SLES 12). On A8 and A9 instances running Windows Server, access to the RDMA network is currently enabled only through applications that use the Microsoft Network Direct interface. See Access to the RDMA network in this article for additional requirements.
A10 and A11 instances have a single, 10-Gbps Ethernet network adapter that connects to Azure services and the Internet.
Azure account – If you want to deploy more than a small number of compute-intensive instances, consider a pay-as-you-go subscription or other purchase options. You can also use your MSDN subscription. See Azure benefit for MSDN subscribers. If you're using an Azure free trial, you can use only a limited number of Azure compute cores.
Cores quota – You might need to increase the cores quota in your Azure subscription from the default of 20 cores per subscription (for Azure Service Management deployments) or 20 cores per region (for Azure Resource Manager deployments), which is not enough for many scenarios with 8-core or 16-core instances. For initial tests, you might consider requesting a quota increase to 100 cores. To do this, open a support ticket at no charge as shown in Understanding Azure limits and increases.
Virtual network – An Azure virtual network is not required to use the compute-intensive instances. However, you may need at least a cloud-based Azure virtual network for many IaaS scenarios, or a site-to-site connection if you need to access on-premises resources such as an application license server. You will need to create a new (regional) virtual network before deploying the instances. Adding an A8, A9, A10, or A11 VM to a virtual network in an affinity group is not supported. For more information, see the virtual network documentation.
Cloud service or availability set – To connect through the RDMA network, the A8 and A9 instances must be deployed in the same cloud service (for IaaS scenarios with Linux-based VMs or Windows-based VMs in Azure Service Management, or PaaS scenarios with Windows Server) or the same availability set (for Linux-based VMs or Windows-based VMs in Azure Resource Manager).
HPC Pack is Microsoft’s free HPC cluster and job management solution for Windows. Starting with HPC Pack 2012 R2 Update 2, HPC Pack supports several Linux distributions to run on compute nodes deployed in Azure VMs, managed by a Windows Server head node. With the latest release of HPC Pack, you can deploy a Linux-based cluster that can run MPI applications that access the RDMA network in Azure. For more information, see the Get started with Linux compute nodes in an HPC Pack cluster in Azure.
HPC Pack is not required for you to use the A8, A9, A10, and A11 instances with Windows Server, but it is a recommended tool to create Windows HPC Server–based clusters in Azure. In the case of A8 and A9 instances, HPC Pack is the most efficient way to run Windows-based MPI applications that access the RDMA network in Azure. HPC Pack includes a runtime environment for the Microsoft implementation of the Message Passing Interface for Windows.
For more information and checklists to use the compute-intensive instances with HPC Pack on Windows Server, see Set up a Windows RDMA cluster with HPC Pack to run MPI applications.
Within a single cloud service or an availability set, the A8 and A9 instances can access the RDMA network in Azure for running MPI applications that use the Linux RDMA drivers to communicate between instances. At this time, Azure Linux RDMA is supported only with Intel MPI Library 5.
Currently, Azure Linux RDMA drivers are not available for installation via driver extensions. They are available only by using the RDMA-enabled SLES 12 image from the Azure Marketplace.
See the following table for prerequisites for Linux MPI applications to access the RDMA network in clusters of compute nodes (IaaS). See Set up a Linux RDMA cluster to run MPI applications for deployment options and configuration steps.
|Prerequisite||Virtual machines (IaaS)|
|Operating system||SLES 12 HPC image from the Azure Marketplace|
|MPI||Intel MPI Library 5|
Within a single cloud service or availability set, the A8 and A9 instances can access the RDMA network in Azure for running MPI applications that use the Microsoft Network Direct interface to communicate between instances. The A10 and A11 instances do not include access to the RDMA network.
See the following table for prerequisites for MPI applications to access the RDMA network in virtual machine (IaaS) and cloud service (PaaS) deployments of the A8 or A9 instances. For typical deployment scenarios, see Set up a Windows RDMA cluster with HPC Pack to run MPI applications.
|Prerequisite||Virtual machines (IaaS)||Cloud services (PaaS)|
|Operating system||Windows Server 2012 R2 or Windows Server 2012||Windows Server 2012 R2, Windows Server 2012, or Windows Server 2008 R2 Guest OS family|
|MPI||MS-MPI 2012 R2 or later, either stand-alone or installed via HPC Pack 2012 R2 or later
Intel MPI Library 5
|MS-MPI 2012 R2 or later, installed via HPC Pack 2012 R2 or later
Intel MPI Library 5
For IaaS scenarios, the HpcVmDrivers extension must be added to the VMs to install Windows network device drivers that are needed for RDMA connectivity. Depending on your deployment method, the HpcVmDrivers extension might by added to a size A8 or A9 VM automatically, or you might need to add it yourself. To add the extension, see Manage VM extensions.
Pricing - The A8–A11 VM sizes are available only in the Standard pricing tier.
Resizing – You can't resize an instance of size other than A8–A11 to one of the compute-intensive instance sizes (A8-11), and you can’t resize a compute-intensive instance to a non-compute-intensive size. This is because of the specialized hardware and the performance optimizations that are specific to compute-intensive instances.
RDMA network address space - The RDMA network in Azure reserves the address space 172.16.0.0/12. If you plan to run MPI applications on A8 and A9 instances in an Azure virtual network, make sure that the virtual network address space does not overlap the RDMA network.