One-to-Many Qualities Central to Success
click to enlarge
The concept of ‘one-to-many’ aptly describes a desire of HPC cluster users, developers and builders that has been central to the success of the volume PC industry. In particular, two qualities stand out
• Starting at the top of the stack and looking down through it, different ISVs register applications that run using the API. These binaries need to be checked and validated by their developers once on a single binary implementation in order to provide high confidence that one binary will run on many implementations.
• Similarly, looking up through the stack, the certified cluster platform providers develop and certify a hardware and system software cluster configuration. These integrators then are free to build implementations of that configuration with a high confidence that many different ISV codes will run on their implementations without needing to verify that specific code on each ISV port.
Note that these two qualities are provided by the components that make up the stack architecture. Each ISV is able to supply additional components, such as an alternate MPI runtime. However, if they do, they will need to provide that component and will need to certify and stand behind its correctness in the stack.
Clement T. Cole is Intel Cluster Ready Architect in Intel’s Software & Services Group. He may be reached at editor@ScientificComputing.com.