Johan De Keyser

Run-time load balancing support for a parallel multiblock Euler/Navier-Stokes code with adaptive refinement on distributed memory computers

Abstract This paper describes the parallel implementation of algorithms requiring run-time load redistribution with the aid of the parallel programming library loco . As a typical application, a 2D finite volume multiblock Euler/Navier-Stokes code with block-wise adaptive mesh refinement is discussed. The loco software handles the communication between blocks and the distribution of blocks among the processors, thereby performing automatic load balancing at run-time. The loco library is interfaced with both the native NX communication primitives on Intel iPSC hypercubes and the PVM software on workstation clusters. The parallel performance of the code on the Intel iPSC/860 and on a DEC Alpha workstation cluster is discussed. In particular the effects of mesh refinement on the load balance are investigated.

Distributed Mapping of SPMD Programs with a Generalized Kernighan-Lin Heuristic

Mapping a parallel algorithm onto a parallel computer is known to be NP-hard. In this paper we propose a parallel mapping technique based on a generalization of the Kernighan-Lin heuristic. The properties of a mapping cost function are thereby exploited.

A Survey of Field-Aligned Mach Number and Plasma Beta in the Solar Wind

We have surveyed solar wind plasma beta and field-aligned Alfvenic Mach number using Ulysses and Wind data. We show the characteristic timescale and occurrence frequency of ‘magnetically dominated’ solar wind, whose interaction with a planetary magnetosphere may produce a bow shock with multiple shock fronts. We discuss radial, latitudinal, and solar cycle effects.

A Parallel Multiblock Euler/Navier-Stokes Solver on a Cluster of Workstations Using PVM

An adaptive 2D multiblock Euler/Navier-Stokes solver has been parallelised using the LOCO software library which performs automatic load-balancing at run-time by an appropriate distribution of the blocks onto the processors of the parallel computer. We describe our experience in porting the solver onto a Unix workstation cluster environment equipped with the Parallel Virtual Machine communication software.

Solar illumination control of the polar wind

Polar wind outflow is an important process through which the ionosphere supplies plasma to the magnetosphere. The main source of energy driving the polar wind is solar illumination of the ionosphere. As a result, many studies have found a relation between polar wind flux densities and solar EUV intensity, but less is known about their relation to the solar zenith angle at the ionospheric origin, certainly at higher altitudes. The low energy of the outflowing particles and spacecraft charging means it is very difficult to measure the polar wind at high altitudes. We take advantage of an alternative method that allows estimations of the polar wind flux densities far in the lobes. We analyze measurements made by the Cluster spacecraft at altitudes from 4 up to 20 RE. We observe a strong dependence on the solar zenith angle in the ion flux density and see that both the ion velocity and density exhibit a solar zenith angle dependence as well. We also find a seasonal variation of the flux density.

A parallel block-structured euler/navier-stokes code with adaptive refinement and run-time load balancing on the iPSC/860

Publisher Summary This chapter describes the parallel implementation of a 2D finite volume multiblock Euler/Navier–Stokes code. The sequential multiblock solver on which the parallel code is based has two sets of arrays containing the mesh data. One set describes the inner part of a block and the other one describes the block border. One of the indices in each array is the block number, i.e., the name of a block is directly related to its position in the memory. In the parallel version, the position of a block in the distributed memory (i.e. the processor number and the position in the memory of that processor) is independent of the name of the block. Every block is given a globally unique name. The coupling between the name on one hand and processor number and position in that processors memory on the other hand is maintained by a global name table provided by LOCO. This table is automatically updated whenever a block is moved to another processor. Communication and load balancing is handled by the parallel programming library LOCO. The parallel performance of the code on the Intel iPSC/860 is discussed in the chapter.