An Outlook of High Performance Computing Infrastructures for Scientific Computing

An overview of the high performance computing (HPC) infrastructures, comprising both the hardware and the software, is presented as a concise introduction for computational scientists. The scientists and engineers who are not well-acquainted with HPC concepts and technologies are the intended audience, in especial, to make them appreciate what computing innovations have reached their doorsteps to offer them the potential that can boost productivity. Especially, the advent and ubiquitous acceptance of multi/many-core CPU chips has brought a sea change that is compelling the scientists and engineers for a paradigm shift so as to perceive and develop the computer solutions accordingly. A variety of well-established and matured trends and technologies for HPC are discussed. Parallelism, which is the most commonly recognized approach of HPC, is firmly categorized into two forms: implicit and explicit (from a programmer's perspective depending upon how much effort is required from the programmer to obtain a certain form of parallelism). Although the term implicit parallelism has been used in research articles for a few years due to the recent advancements in the microprocessors and compilers but such a categorization is somewhat lacking in textbooks. Further, the notion of instruction level parallelism (ILP) is discussed under implicit parallelism. As it is very important to adopt efficient programming approaches to get maximum benefit of an available HPC infrastructure, a separate section on the development of efficient parallel programs is also presented.