Open Grid Forum (OGF) recently achieved a major leap forward in ensuring grid interoperability becomes real. Microsoft, Altair Engineering and Platform Computing have adopted OGF’s High Performance Computing Basic Profile Specification in next-generation product releases. The HPC Basic Profile Specification incorporates two OGF published standards –В the OGSA-Basic Execution Services specification and the Job Submission Description Language specification — together with the WS/I Basic Profile published by the Web Services Interoperability Organization.
Starting in 2008, Microsoft’s Window’s HPC Server 2008, Altair’s PBS Professional and Platform’s LSF products all will meet OGF’s HPC Basic Profile Specification. End users will now be guaranteed that their use of these products will enable complete interoperability to manage resources in virtual organizations over multi-site, multi-vendor grids. Furthermore, end users will be able to integrate their use of these highly function commercial solutions together with open source solutions developed by the worlds’ leading production grids.
OGF, Microsoft, Altair and Platform, together with EGEE/OMII-UK, NorduGrid/KnowARC, NIC/Forschungszentrum Julich/OMII Europe, UK e-Science and the University of Virginia, demonstrated fully interoperable implementations of the HPC Basic Profile at SC 2007. Demonstrations involved compute clusters on a grid processing various applications submitted via the HPC Basic Profile specification. HPC Basic leveraging common Web services and OGF standards to ensure all middleware used in the demonstration worked seamlessly together.
End users will benefit from the integration of HPC Basic Profile into multiple applications in a variety of ways:
- Web application: Users access applications with their Web browser while the application server uses theВ HPCВ Basic Profile to initiate the execution of the application.В This capability, for example, has provided engineers working in oil reservoir exploration an easy to use interface for compute job submission and execution tracking, regardless of which compute device is being used.
- Metascheduler: Users submits their jobs to a local resource manager, which then uses the defined policy to route the jobs to a suitable execution resource. This resource may be accessed through theВ HPCВ Basic Profile and use a different resource manager or be located in a different organization.В This capability simplifies the consolidation of separate grids (perhaps due to a merger or partner arrangement) which may be running different scheduling software.
- Rich client:В An engineer or scientist runs a “workbench” application that includes running simulations that test elements of the design. When the user runs a simulation, the rich client dispatches it to a compute resource using theВ HPCВ Basic Profile. In many cases, this has allowed organizations to gain better leverage of their entire set of compute devicesВ – regardless of geographic location — resulting in higher utilization, lower cost of operation and better return on capital investment.
- Workflow engine:В Various applications are chained together into a workflow of computing tasks. Users define job nodes and their dependencies into a graph. The workflow engine does not need to be aware of the different resource managers and can execute each node in the graph based on the job description and theВ HPCВ Basic Profile.В This capability saves time and reduces the risk of human intervention during the execution of a multi-stage workflow such as one that a crash simulation engineer may require.