This post has been republished via RSS; it originally appeared at: Azure Compute Blog articles.
Announcing Univa Grid Engine Cluster for Azure CycleCloud
Univa Grid Engine enterprise-grade workload management including first-class support for container workloads meets Azure CycleCloud is the simplest way to manage high-performance computing (HPC) workloads. This collaboration realizes end-to-end enterprise support for HPC workloads running on Azure – including a free, preconfigured demo experience to explore and learn more about the capabilities of this combined offering. In Azure Cyclecloud 8 you’ll find a Univa Grid Engine demo cluster with rich autoscaling capabilities – ready to dive into.
Full Grid Engine and Azure capabilities
Grid scalability has been a consistent focus for both Univa and Azure, with this offering customers will benefit from scheduling efficiencies and elasticity – graduate to a fully-licensed version of Grid Engine to unlock cloud computation and increase its speed on a staggering scale. The solution provides extraordinary scale, and precise resource allocation at scale. Close collaboration brings resource-awareness throughout the stack – specify 4 NVIDIA Tesla V100 GPUs in the job requirements and 4 GPUs will be added to the grid, or 400 or 4000.
Multi-dimensional autoscaling based on resources
Introduced as part of Cyclecloud 8 is multi-dimensional autoscaling. Complex job requirements such as memory, CPUs, GPUs will all be evaluated, and the grid will scale to meet all requirements for all jobs – in a minimal fashion. You get everything you need and no more than you need for the workload. Grid Engine jobs with resource requirements will translate to real grid resources through Cyclecloud. Also available are tunable scale-down parameters to reduce size as resource become idle. You can toggle between fill-up and round robin resource autoscaling logic to be more cost-conservative or performance-driven. This resource allocation is distinct from job scheduling – where all the rich scheduling resources provided by Grid Engine are available. The combined solution is a resource-aware highly elastic HPC compute environment.
Azure is the best place to run HPC workloads
This solution brings first-class support for Azure Infiniband networking capabilities. Dedicate specific Grid Engine hostgroups for specific Infiniband fabrics so that tightly coupled jobs run on the most performant network – all automatically when you use the mpi parallel environment (or configure your own Grid Engine PE to use IB).
The cluster-side component is an installable package such that an existing Grid Engine cluster can be enhanced with Azure VMs. This hybrid scenario can extend your datacenter with auto-scaling Azure HPC VMs.
Unlock Univa Grid Engine on Azure
Contact Univa to upgrade from the preconfigured demo version cluster to an unlocked version of Grid Engine to reach enterprise scale and support. The Cyclecloud cluster is ready to run with a pre-packaged demo license of Univa Grid Engine.
Find More Information About Grid Engine