Industrial Applications of High-Performance Computing

xvii

Executive Summary

T

         A

following SURFsara’s Superdag annual event for e Netherlands

HPC community to get informed on recent developments and oppor-

tunities, and get in touch with each other. Superdag is specically orga-

nized for users of the national ICT (Information, Communication, and

Technology) research infrastructure at SURFsara. At this particular meet-

ing, Prof. Anwar Osseyran invited Merle Giles from NCSA to speak about

why we as a community should care about industrial supercomputing.

e dinner request to collaborate on a book seemed a natural one, as the

authors had met previously as part of an informal group made up of super-

computer centers that serve and collaborate with industrial users.

is informal group was named International Industrial Super-

computing Workshop, an appropriately descriptive tag rst used by

Dr. Sang Min Lee at KISTI Supercomputer Center in Daejeon, South

Korea, in 2010. e second workshop was hosted by Dr. Michael Resch at

HLRS in Stuttgart, Germany, the home of world-class collaboration with

automobile manufacturers Mercedes Benz and Porsche. e third was

hosted at NCSA at the University of Illinois, the fourth at SURFsara in

Amsterdam, and the h by University of Tokyo in Japan.

e rules were simple: attendees must come from supercomputer centers

that serve industrial users, and vendors were not allowed. Typical super-

computer meetings are rather heavy on technology, as this business is fast-

moving and very innovative. Industry, in this case, was dened as end-users

of supercomputing technology rather than the more common reference of

supercomputing as a provider industry. e common bonds of this group

were rooted in the customer service models and unique challenges of serv-

ing users from industry rather than university and academic researchers. A

fundamental dierence between these user communities is that academic

researchers obtain allocations on supercomputers as part of peer-reviewed

xviii ◾ Executive Summary

grant processes, and they run applications on these machines at no cost.

Industry, on the other hand, regularly pays for such use.

Historically, supercomputers are funded primarily by (1) national gov-

ernments, (2) universities, and (3) individual companies. ese resources

dier mostly in terms of scale, with the largest computers funded mostly by

nations, mid-tier machines by universities, and mid-tier to entry-level com-

puters by individual companies. Social and economic benets occur along

all scales, but the best-known and most publicized benets are from aca-

demic communities, which by virtue of receiving no-cost allocations must

publish their results for the greater good. Industry use ranges from pure

commercial and production purposes to precompetitive and open scientic

use, with the latter requiring the sharing and publishing of results.

We attempt to tell a story about how industry demand for supercom-

puting is rooted in science and engineering that is similar to that of aca-

demic science and engineering. Dierences occur in the scale of discovery

and knowledge, with academic scientists seeking scientic insight, whereas

industrial scientists and engineers seek insight that manifests in products

and services, typically on a shorter time-to-discovery scale. Big science can

use large fractions of supercomputers, whereas engineering typically uses

smaller fractions. Academic research allocations on supercomputers may

lead to publishable discoveries with no guarantee of subsequent allocations,

whereas industrial use tends to run iterations of simulations to seek an opti-

mal solution. e former can mostly be described as capability computing

when the researcher makes optimal use of the machine capabilities to solve

one large complex problem; the latter is oen described as capacity com-

puting. We should point out that scientists in both academia and industry

require capability computing for scientic insight and that both communi-

ties utilize capacity computing routinely. e distinction here is that gov-

ernment policies around the world oen favor capability computing in the

pursuit of science; private-sector computers favor optimization and capacity

computing. In all cases, the terms supercomputing and high-performance

computing, or HPC, will be used interchangeably. e Supercomputing 101

introduction will help nontechnical readers understand the fundamental

aspects of how HPC diers from normal, everyday computing.

is book is not intended to be a comprehensive review of the global

state of supercomputing. It does attempt to give the reader a global sense

of technological purpose. Provisioning vast technical resources, operat-

ing these systems, programming them, and taming these ckle beasts

take immense resources for personnel. Because they are not mass-market