insideHPC Guide to Successful Technical Computing

Many scientists and engineers have found that successful technical computing can solve problems, lead to insights, and generate discoveries faster and more efficiently than any other method-often reducing solution times from days or weeks to hours.

A practical approach to HPC assumes clearly defined goals that prevent the solution from becoming its own research project. An HPC solution should allow users to focus on their research and not the nuances of the HPC system. Non-HPC users are often surprised to learn that HPC delivers a competitive advantage and remarkable return on investment. HPC has gone from a specialized back-room art to an essential and competitive tool. Some of the areas that benefit from HPC include Higher Education, Life Sciences, Manufacturing, Government Labs, Oil and Gas, and Weather Modeling and Prediction.

The use of commodity processors had helped bring HPC to the masses. Modern processors, like those from Intel, now employ multiple cores and offer exceptional value. Scaling these processors to attack larger problems can take two forms. The first and most effective method scales up the number of processors using a large pool of shared memory.

The second is a clustered scale-out approach where multiple separate servers are combined using highspeed networks.  The scale-up approach has demonstrated success in many areas. In particular, certain bioinformatics problems run exceptionally well on scale-up systems but are not tractable with scale-out designs. The simplicity of scale-up systems also offers advantages with large multiuser systems usually found in academic environments. Similar success has been found in other industries.

The growth of cloudbased computing has become very popular. Before a cloud solution is considered, however, the cost and time of data movement
should be determined. In many situations, the location of the data may determine where the computation takes place.  This paper offers those considering HPC, both users and managers, guidance when considering the best way to deploy an HPC solution. Three important questions are suggested that help determine the most appropriate HPC design (scale-up or scale out) that meets your goal and accelerates your discoveries.