HPC containers with Azure Batch

With the latest updates to Azure Batch, you now have the option to schedule your tasks as Docker container invocations. Containers and Azure Batch are an ideal way to package, execute, and scale your High Performance Computing (HPC) applications and batch workloads in a consistent, reproducible manner utilizing powerful cloud native job scheduling capabilities.

Today, we're excited to announce support for Singularity containers in the latest Batch Shipyard release. Singularity is a container solution amenable to both administrators and users of shared HPC and cluster computing environments, while still providing access to accelerators such as GPUs and specialized interconnects in container contexts. Batch Shipyard is an open system for enabling simple, configuration-based container execution on Azure Batch, and aims to allow users of these shared computing environments to easily execute their existing Singularity workloads on Azure. Azure's GPU, including ND, NCv2, and the upcoming NCv3 series of VMs, and RDMA-enabled instances are potentially an ideal fit for such workloads. Also, with Batch Shipyard and Azure Batch, not only can you automatically scale your compute pools with ease, you can also opt to execute your workloads on low priority VMs for savings up to 80%!

In addition to support for Singularity, the latest release of Batch Shipyard includes preliminary support for Windows containers, integrated Azure Batch container support, YAML based configuration support, ability to reference multiple private registries, ARM Image-based custom images, and pre-built binaries for the CLI among other improvements. Please view the Change Log and the migration guide if you are upgrading from a previous version. If desired, Batch Shipyard is available directly within Azure Cloud Shell with no installation required.

Source: Azure Blog Feed

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.