Advanced Computing

255 Pages·2013·5.386 MB·English

by  Dominik Schillinger, Quanji Cai, Ralf-Peter Mundani, Ernst Rank (auth.), Michael Bader, Hans-Joachim Bungartz, Tobias Weinzierl (eds.)

93
Michael Bader · Hans-Joachim Bungartz
Tobias Weinzierl Editors
Advanced
Computing
Editorial Board
T. J.Barth
M.Griebel
D.E.Keyes
R.M.Nieminen
D.Roose
T.Schlick
Lecture Notes
in Computational Science
93
and Engineering
Editors:
TimothyJ. Barth
MichaelGriebel
DavidE. Keyes
RistoM. Nieminen
DirkRoose
TamarSchlick
Forfurthervolumes:
http://www.springer.com/series/3527

Michael Bader • Hans-Joachim Bungartz
Tobias Weinzierl
Editors
Advanced Computing
123
Editors
MichaelBader
Hans-JoachimBungartz
TobiasWeinzierl
DepartmentofInformatics
TechnischeUniversitätMünchen
Munich,Germany
ISSN1439-7358
ISBN978-3-642-38761-6 ISBN978-3-642-38762-3(eBook)
DOI10.1007/978-3-642-38762-3
SpringerHeidelbergNewYorkDordrechtLondon
LibraryofCongressControlNumber:2013950610
Math.Subj.Class.(2010):35-06,68U20,68W10,65Y05,65K05,65Y20,76D05,76S05,68N30,68U05
©Springer-VerlagBerlinHeidelberg2013
Thisworkissubjecttocopyright.AllrightsarereservedbythePublisher,whetherthewholeorpartof
thematerialisconcerned,specificallytherightsoftranslation,reprinting,reuseofillustrations,recitation,
broadcasting,reproductiononmicrofilmsorinanyotherphysicalway,andtransmissionorinformation
storageandretrieval,electronicadaptation,computersoftware,orbysimilarordissimilarmethodology
nowknownorhereafterdeveloped.Exemptedfromthislegalreservationarebriefexcerptsinconnection
with reviews or scholarly analysis or material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work. Duplication of
this publication or parts thereof is permitted only under the provisions of the Copyright Law of the
Publisher’slocation,initscurrentversion,andpermissionforusemustalwaysbeobtainedfromSpringer.
PermissionsforusemaybeobtainedthroughRightsLinkattheCopyrightClearanceCenter.Violations
areliabletoprosecutionundertherespectiveCopyrightLaw.
Theuseofgeneraldescriptivenames,registerednames,trademarks,servicemarks,etc.inthispublication
doesnotimply,evenintheabsenceofaspecificstatement,thatsuchnamesareexemptfromtherelevant
protectivelawsandregulationsandthereforefreeforgeneraluse.
While the advice and information in this book are believed to be true and accurate at the date of
publication,neithertheauthorsnortheeditorsnorthepublishercanacceptanylegalresponsibilityfor
anyerrorsoromissionsthatmaybemade.Thepublishermakesnowarranty,expressorimplied,with
respecttothematerialcontainedherein.
Printedonacid-freepaper
SpringerispartofSpringerScience+BusinessMedia(www.springer.com)
Themostscientificallyimportantand
economicallypromisingresearch frontiersin
the21stcenturywillbeconquered bythose
mostskilledwithadvanced computing
technologiesandcomputationalscience
applications.
From:“ComputationalScience:EnsuringAmerica’s
Competitiveness”,President’sInformationTechnology
AdvisoryCommittee(PITAC),2005

Preface
Advanced Computing
Theopeningstatementofthisbook,oneofthekeyconclusionsofthe2005PITAC
report,labelsAdvancedComputingandcomputationalscienceapplicationsasfields
of research with high scientific and economic relevance. Today, computational
scienceandengineering(CSE)hasbecomeahighlyinterdisciplinary,challenging,
and thriving scientific domain of increasing importance and visibility in research,
development,andeducation.Itsessenceinvolvesmodellingandcomputations–in
particular (numerical) simulations – on computer systems ranging from standard
PCtosupercomputers.Thus,CSEapplicationshavealwaysbeenadrivingforceof
computingtechnologies,especiallyprogressingthefieldofsupercomputing.
However, regarding advanced computing technologies, an increasing gap is
observedbetweenwhatshouldbepossibleintheory–duetorecentadvancements
inalgorithms,hardware,andnetworks–andwhatcanreallybeachievedinpractice.
While increase of hardware performance (Moore’s law) and considerations on
complexityandaccuracyofalgorithmsweredrivingthisfieldoverthelastdecades,
new limiting factors are now encountered: hardware awareness and ubiquitous
parallelism due to many-core systems (to name only two such issues) transform
computinginto a multifaceted process spawning modelling, algorithmics, parallel
programming,andevenhardwaredesign;simulationdataareproducedtoanextent
andresolutionwhichmakestheirmerehandling,theiranalysisand,thus,theextrac-
tion and representation of relevant information a serious challenge – especially
if to be done interactively; the development of software for high-performance
computing (HPC), ranging from specialised single-application codes to general
problem-solvingenvironments,hasthusbecomeanincreasinglycomplexprocess,
comparable to what we have known for decades from other fields, and it needs
sophisticatedlanguageandtoolsupportaswellasasubstantialprofessionalisation.
Hence, to be able to exploit future (super)computersat their full potential and
toexpandtheresearchfrontiersmentionedabove,aconcertedeffortisnecessaryin
AdvancedComputing–whichisthusdefinedasHPCtogetherwithallitsenabling
vii
viii Preface
technologies crucial to tackle the old and new bottlenecks. Such an endeavour
must combine computing technologies and applications and bring together the
HPC community as well as application specialists with experts from the enabling
technologies.
TheMunichCentre of Advanced Computing
Starting from these considerations, in 2008, the Munich Centre of Advanced
Computing (MAC) has been established as an independent research consortium
andcommonroofforcomputing-relatedactivitiesofresearchgroupsatTechnische
Universität München (TUM) – together with several partner institutions, such
astheLeibnizSupercomputingCentre,theMax-PlanckInstituteforAstrophysics,
ortheDepartmentofGeophysicsofLudwig-Maximilians-UniversitätMünchen.On
thefundingside,MACstartedontwopillars:eightprojectscofinancedbytheState
ofBavariaandbyTUMandtwoprojectsestablishedintheframeworkofastrategic
partnershipbetween TUM and KAUST, the King Abdullah University of Science
andTechnology,inJeddah,SaudiArabia.FosteringastructuredprogrammeforPhD
studies, MAC is participatingin TUM’s InternationalGraduateSchoolof Science
andEngineering(IGSSE).AllMACPhDstudentsthusengagedinIGSSE’sresearch
trainingprogramme–oneofthecornerstonesbeinga3-monthresearchvisitatan
internationalpartnerinstitute.
Articles inTheseProceedings
Thepresentproceedingsvolumecombines11articlesthatreviewresearchworkand
resultsfromtheMACprojectteams.Thefirsttwoarticles,bySchillingeretal.and
Benketal.focusontheFinite CellMethodandonImmersedBoundaryMethods,
respectively, and thus on the question how to efficiently and accurately discretise
problems with complicated domain boundaries on structured (adaptive) Cartesian
grids,inordertosimplifymeshgenerationandimprovecomputationalperformance.
Cai et al. discuss how micro- and macroscale modelling can be combined for
accuratesimulation of porousmedia flow,with CO -sequestrationas the intended
2
application.SimonandUlbrichpresenttheirworkonoptimalcontrolproblemsfor
thisapplicationarea.
As an example for the development of novel mathematical algorithms, Böhm
and Ulbrich present a Newton-CG method for full-waveform seismic inversion.
Roderus et al. demonstrate how advancing both mathematical and computational
algorithmscan lead to substantial improvementof the performanceof established
computational science codes on HPC platforms – their focus being on density
functionaltheoryasapplication.Krajaetal.intheirarticle,puttheHPCplatforms
first in the development chain and discuss how to design HPC architectures for
Preface ix
given applications; here, on-board processing for SAR image reconstruction on
spacecraftsisconsidered.Assimulationtasksandcodes,respectively,becomemore
and more complex, systematic software engineering has become imperative – Li
etal.addressthisquestionviaarequirementsengineeringapproachforasoftware
packageincomputationalseismology.
The following two articles both address interactive simulations: while by
Kneževic´ et al. present an interactive computing framework for engineering
applications,inparticular,Benzinaetal.introduceaframeworkthatusessurrogate
models based on sparse grids to allow interactive handling of high-dimensional
simulation data. Both projects used the FRAVE – a flexible, reconfigurable
visualisation environment installed at MAC – for their studies. Tönnis et al. as
our last article in these proceedings, focus on the developers and users of such
interactivesimulationandvisualisationenvironments:theypresentexperiencesand
findingsregardingergonomicaspectsofsettingupandworkingwiththeFRAVE.
Takenaltogether,thearticlesthusprovideanoverviewofresearchinMACfrom
2008to2012andcover(nearly)theentiresimulationpipeline.
Munich,Germany MichaelBader
May2013 Hans-JoachimBungartz
TobiasWeinzierl