Optimised grid partitioning for block structured grids in parallel computing [Elektronische Ressource] / von Daniel Junglas

Optimised Grid-Partitioning for BlockStructured Grids in Parallel ComputationVom Fachbereich Mathematikder Technischen Universit¨at Darmstadtzur Erlangung des Grades einesDoktors der Naturwissenschaften(Dr. rer. nat.)genehmigte DissertationvonDipl.-Math. Daniel Junglasaus Freiburg im BreisgauReferent: Prof. Dr. A. MartinKoreferent: Prof. Dr. M. Sch¨aferTag der Einreichung: 27. April 2006Tag der mu¨ndlichen Pru¨fung: 12. Dezember 2006Darmstadt 2007D17AbstractSimulation of turbulent flows in complex geometries is nowadays usually performed by ap-plication of grid-based algorithms on parallel computers. In this approach it is not onlyimportant to have a clever discretisation and an appropriate grid. One must also use (ordevelop) algorithms that are convergent and numerically stable. As a final ingredient tosuccess the different work packages of the simulation must be distributed over the set ofavailable processors. This distribution must be performed so as to optimally exploit thecomputational resources provided by processors, thereby minimising simulation time.Our work will focus on this last optimisation problem and develop models as well solutionalgorithms forit. To this end we assume that a block structured as well as a suitable simula-tionalgorithmarefixedandlookforanoptimalmapping ofblockstoprocessors.