This post has been republished via RSS; it originally appeared at: Azure Compute Blog articles.
As Azure continues to deliver new H-series and N-series virtual machines that match the capabilities of supercomputing facilities around the world, we have at the same time made significant updates to Azure CycleCloud -- the orchestration engine that enables our customers to build HPC environments using these virtual machine families.
We are pleased to announce the general availability of Cyclecloud 8.1, the first release of CycleCloud on the 8.x platform. Here are some of the new features in this release that we are excited to share with you.
Welcome Univa Grid Engine
Whether they are tightly-coupled or high-throughput, submit your jobs to your scheduler of choice and CycleCloud will allocate the required compute infrastructure on Azure to meet resource demands, and automatically scale down the infrastructure after the jobs complete. With this addition, CycleCloud is the only cloud HPC orchestration engine that supports all the major HPC batch schedulers, one that you can rely upon to run your production workloads.
Configure Shared Filesystems Easily in the UI
In CycleCloud 8.1, it is simple to attach and mount network attached storage file systems to your HPC cluster. High-performance storage is a crucial component of HPC systems; Azure provides different storage services, such as HPC Cache, NetApp Files, and Azure Blob, that you may use as part of your HPC deployment, with each service optimized for different workloads.
Most workloads typically require access to one or more shared filesystems for common directory or file access, and you can now configure your cluster to mount NFS shares through the CycleCloud web user interface without needing to modify configuration files.
Perform Last-Mile VM Customization with Cloud-Init
CycleCloud Projects is a powerful tooling system for customizing your cluster nodes as they boot-up. In CycleCloud 8, we have included support for a complementary approach with Cloud-Init, a widely-used initialization system for customizing virtual machines.
You can now use the CycleCloud UI or a cluster template to specify a cloud-init config, Shell script or Python script that will configure a VM to your requirements before any of CycleCloud's preparation steps.
Trigger Actions through Azure Event Grid
CycleCloud 8 generates events when certain node or cluster changes occur, and you can configure your CycleCloud installation to have these events published to Azure Event Grid. With this, you can create triggers for events like Spot VM evictions or node allocation failures, and automatically drive downstream actions or alerts . For example, track Spot eviction rates for your clusters by VM SKU and Region, and use those to trigger a decision to switch to pay-as-you-go VMs.
Azure is the Best Place to Run HPC Workloads
Over the past year, we have been privileged in being able to help our partners run their mission critical HPC workloads on Azure. CycleCloud has been used by the U.S Navy to run weather simulations for forecasting cyclone paths and intensities, by structural biologists to obtain one of the first glimpses at the crucial SARS-Cov-2 Spike protein, by biomedical engineers in performing ventilator airflow simulations ahead of an FDA EUA submission.
With the CycleCloud 8.1 release, we look forward to continuing our support of the HPC community and partners by giving them familiar and easy access to scalable, true supercomputing infrastructure Azure.