IBM has built the tools and resources required to pull together silos of HPC into a single, allocatable entity
IBM is preparing to raise high performance computing (HPC) into the cloud. The company has announced plans to help customers link disparate computing resources to form a single, large, private cloud supercomputing environment.
HPC resources are often deployed to support a single workload or project. IBM argues that the result is a series of silos with dedicated resources and no possibility of supporting other workloads.
Combining The Silos Of Computation
By providing the tools required to manage an HPC infrastructure using cloud computing principles, the silos can be dispelled and resources can be allocated on a project-by-project basis. This maximises the operational efficiency of existing computational clusters by creating a pool of resources.
IBM’s offering can enable customers to manage and prioritise HPC assets on a global basis. This is particularly applicable to multinational operations, such as the large oil and gas companies, manufacturing enterprises, weather and environmental prediction bureaux and scientific research centres, that need to pool large amounts of data for analysis processes.
In a conventional HPC silo, the loads fluctuate according to the stage of processing reached so an expensive supercomputing installation could be virtually idle. In IBM’s cloud environment the redundant elements of the cluster could be reallocated to divisions or departments that need extra power.
This not only offers the ability to pool resources but can also allow an organisation to save money by reducing the overall computational environment or to step up its effective capacity through careful management.
The process is similar to the management capabilities offered by specialist companies such as Platform Computing. Red Bull Racing, the current leading Formula 1 team, uses Platform to allocate the resources of its HPC facility. The software also allows the F1 outfit to add extra power by harnessing unused desktop workstations when the designers and office staff finish working for the day.
The HPC cloud offerings from IBM include:
- HPC Cloud Implementation Service from IBM: A quick-start service to help install, configure and optimise a private cloud or private-hosted cloud combination. This includes system administrator training;
- IBM HPC Management Suite for Cloud: A resource manager with tools to enable administrators to manage the cloud, schedule jobs and handle tasks like metering energy usage while allowing end users to view and access the full set of HPC resources available to them in the cloud;
- IBM Intelligent Cluster: An integrated, optimised HPC cluster. The servers, storage and switches are factory-integrated, tested and delivered as a plug-and-go addition for a data centre.
These products and IBM’s Engineering Solutions for Cloud will be available this autumn.