Solution architecture: HPC cluster deployed in the cloud
High-performance computing (HPC) applications can scale to thousands of compute cores, extend on-premises big compute or run as a 100% cloud-native solution. This HPC solution, including the head node, compute nodes and storage nodes, runs in Azure with no hardware infrastructure to maintain.
This solution is built on the Azure-managed services: Virtual Machine Scale Sets, Virtual Network and Storage. These services run in a high-availability environment that is patched and supported, allowing you to focus on your solution instead of the environment they run in.
Implementation guidance
| Products/Description | Documentation | |
|---|---|---|
|
HPC head node |
The HPC head node runs on premises or within Azure as a virtual machine. Size A8/A9 Azure virtual machines provide HPC compute nodes running on Windows or Linux operating systems. RDMA networking, available with A8 and A9 instances, is used to achieve high bandwidth and microsecond latencies between compute nodes. Storage nodes can also be run within virtual machines. |
|
|
Virtual Machine Scale Sets |
Availability sets ensure that the application is available and resilient to updates and hardware faults. Virtual machines communicating via RDMA are placed within the same availability set. |
|
|
Virtual Network |
Virtual Network provides IP connectivity between the head node, compute nodes and storage nodes. | |
|
Storage |
Azure Storage blobs store the disks that back the virtual machines and provide long-term storage of unstructured data and executable files used by the HPC application. | |
|
Azure Resource Manager templates |
Resource Manager templates or script files are used to deploy your application to the HPC environment. |