Computer Science Stack Exchange is a question and answer site for students, researchers and practitioners of computer science. It only takes a minute to sign up.
Sign up to join this community
I wonder if high-performance computing and parallel computing always mean the same? If not,
Are high-throughput computing and high-performance computing the same concept?
As noted by @JukkaSuomela, you can do parallel computing on low-end resources such as your laptop and even on your mobile phone (if they are equipped with a multicore processor). However, HPC (High Performance Computing) is, roughly stated, parallel computing on high-end resources, such as small to medium sized clusters (ten to hundreds of nodes) up to supercomputers (thousands of nodes) costing millions of dollars. Therefore, the difference is mainly in the hardware used.
Finally, HTC (High Throughput Computing) refers to executing the maximum number of tasks (or jobs if you prefer) per time unit. A classical example of HTC computation is the so called parameter sweep, in which you must run the same executable, but varying in each execution a set of parameters (thus the sweep spans the whole parameter space).
I would say parallelism is included in HPC. HPC uses subsidiary tools other than paralelism, the most important being vectorization, including matrix operations with diagonalisation and more general SVD speed boost. So HPC wherewithal are often engaged for big problems only well fitted for matrix(Linpack), well conditioned : simulation, PDE. So it ranges from weather predictions, brain simulation, PDE resolution in near real time for High Frequency Trading, and so on. Further applications on www.top500.org
the field of supercomputing has undergone a significant paradigm shift in the last decade or so such that it used to require highly specialized hardware and designs. a supercomputing company that embodied this approach is Cray supercomputers.
however a new concept/trend has emerged that extremely high performance can be achieved with "off-the-shelf" components such as those used in consumer electronics and heavy internetworking using standard networking components. supercomputer designs have always utilized a large amount of parallelism but this trend has been amplified with the proliferation of the off-the-shelf type approach. also parallelism is much more being supported with sophisticated software architectures such as MPI, MapReduce, etc
so the systems are still highly parallel but the particular means of achieving that parallelism has shifted in the hardware and software. the parallelism is more visible at the application layer and not abstracted away.
also recently power consumption has become much more of an issue and supercomputers are starting to focus on energy-efficient designs and low-energy CPUs such as those used in mobile computing. [1] is a nice ref on key issues in current trends/barriers on high performance computing scalability.
[1] Next-Generation Supercomputers Kogge
