This week Platform Computing announced new products aimed at letting cloud resource managers do some fancy footwork with dynamically provisioned images on nodes managed by LSF and Symphony
Platform ISF Adaptive Cluster is a new product offering in the Platform ISF family that enables organizations to create an internal cloud for running HPC applications and is integrated with Platform LSF and Platform Symphony. The product dynamically changes the operating system and associated application stack on demand thereby maximizing resource usage and service levels based on workload. It does this by automatically provisioning Linux and Windows operating systems from bare metal or initiating the right guest operating system within a virtual machine container on the compute nodes upon application demand.
To ensure application availability and guaranteed performance, many enterprises employ multiple HPC clusters, each assigned to run a specific application that requires either a Linux or Windows environment. Inevitably this results in cluster silos and “cluster sprawls” with low utilization and high management costs. Platform ISF Adaptive Cluster consolidates the previous application silos into an internal cloud environment that dynamically provisions Linux and Windows operating systems based on application demand to maximize resource usage and service levels. Utilization rates of separate server clusters typically ranging between 40 – 50 percent – can be increased to 80 – 90 percent.
Other providers offer similar capabilities — Adaptive Computing, for example. This is a feature that is not uncommon in enterprise environments, but it is still in limited use in traditional HPC centers.
The other part of this announcement is some plumbing that will allow customers to redirect workloads out to pay-as-you-go resources outside of a user’s cluster during periods of peak demand
Delivered as a solution in the Platform ISF family, the new cloud bursting capability allows Platform HPC customers to access external cloud resources to offload peak workloads from internal HPC infrastructure on a pay-per-use basis. The new solution allows HPC users to use external resources while operating in a managed, private cloud environment as required by enterprise governance and compliance regulations. Platform is working with external cloud service providers to publish a machine image of the Platform HPC management software stack, which includes Platform LSF, Platform Symphony, Platform Cluster Manager and the new Platform MPI, to ensure Platform customers have a single, user-friendly interface to transparently leverage both their internal enterprise infrastructure for HPC and the external cloud for overflow.