Author: Michael Jackson, President, Adaptive Computing
Welcome to the blog series, “Seven Keys to Success”. During this series we will introduce solutions to the growing complexity of Adaptive Computing’s customer environments and how we applied strategies to take that complexity out of the analysis process and make it manageable, successfully getting the results necessary to advance our customer’s causes. The first key to success is Consolidation.
Consolidation is a key method of breaking down the different siloed or fractured environments that exist in High Performance Computing (HPC). Through the consolidation of physical and virtual resources (grid), an environment can be shared between multiple users and groups, thereby creating a foundation for efficiency.
Homogenous environment are very prevalent, but optimizing application performance on those homogenous environments is not the easiest to thing to do manually. Some applications need more memory, others more I/O, others faster processors, while others can actually run just as well on slower processors. Therefore, using a variety of resources actually optimizes performance as it provides the ideal resource for each application. However, this variety creates a fractured nature within the environment. What’s needed is a workload and resource orchestration manager, like Moab, to reach across heterogeneous resources and allocate workloads and intelligently assign applications to their ideal resource, but then select an alternative resource if the first is already being used, thus maximizing utilization.
This leads us to the next question about consolidation – how do you share resources with departments that previously had their own dedicated resource? Each department wants a guarantee that they’ll get the same amount of resources they would have had in their siloed dedicated environment. We find, that through policy-based workload management with Moab, everyone can still have access to the same amount of resources and, in fact, more, as consolidation allows users to utilize the excess resources. This brings up the concepts of prioritization, reservations, charging, preemption, SLA’s, and quality of service policies. With these policies, organizations can assign different costs to resources and then allow users to select workload importance and apply better service levels to projects, applications, or departments, based on the organization’s priorities.
What if physical resources are separated geographically? Capabilities within grid technology join nearby clusters creating a local area grid, and wide area grids, for clusters that are geographically separated—or that just have a different user space, data space, or administrative domain. Elastic computing which gives HPC administrators the ability to manage resource expansion by bursting to other data center resources utilizing OpenStack or other standard platforms. It helps admins better manage the provisioning and performance challenges of bursty workloads. Balance is given to the environment by dynamically adding resources to an existing cluster or re-purposing servers with different OS’s like Red Hat, CentOS, or SLES.
These are just a few concepts that help administrators consolidate resources and maximize utilization. We can see why Consolidation is the first key to success for administrators to achieve maximum utilization, permit usage-sharing, realize better efficiency, reduce costs, and allow applications to run where they might have been blocked in an otherwise siloed model.