Maria Lucka

Parallel cooperative savings based ant colony optimization - : Multiple search and decomposition approaches

In this paper we study different parallel implementations of the Savings based Ant System algorithm developed for solving the Vehicle Routing Problem. We analyze the effects of low-level parallelization, multiple search strategies and domain decomposition approaches. For the different strategies speedup and efficiency as well as solution quality are reported. Different information exchanges are analyzed within the multiple search strategies.

A Parallel Strongly Implicit Algorithm for Solving of Diffusion Equations

We present a parallel algorithm for the solution of partial differential equations representing a 3-D diffusion process of the underground water by a finite difference method. The algorithm belongs to a class of the incomplete LU factorization methods, where corresponding system of linear algebraic equations is solved by a quasi LU decomposition in every time step. A code realizing the algorithm was written in Fortran 90 programming language using the MPI message passing interface system and was tested on a SGI Origin 2000 parallel computer.

Grid Services for Parallel Molecular Dynamics with NAMD and CHARMM

High-end computing facilities are key to enabling molecular dynamics simulation of biological relevant systems for a sufficiently long simulation period. On the one hand, the trajectory production necessitates a high-performance parallel computer architecture. On the other hand, hundreds of jobs, both trajectory production and analysis, have to be supervised and started more or less automatically. Although high performance computers are now available to a broad scientific community, users are often forced to cope with many low-level details when using these machines for scientific investigations. Grid computing technologies promise to provide seamless access to parallel computers through the abstraction of services hiding the details of the underlying software and hardware infrastructure. In this article, we describe the provision of Grid services for parallel molecular dynamics simulations based on NAMD and CHARMM within the Vienna Grid Environment.

Multiperiod Portfolio Management Using Parallel Interior Point Method

Publisher Summary This chapter describes an algorithm for solving a three-stage stochastic linear program based on the Birge and Qi factorization of a constraint matrix product in the frame of the primaldual path-following interior point method. The algorithm is scalable and enables to solve large linear programs raising from the portfolio management problems. The parallelization of the method for distributed-memory machines using fortran/mpi and the linear algebra package Lapack, is outlined. Parallel implementation of the three stage algorithm in the frame of the parallel interior point algorithm (IPM) method is based on the message passing interface (MPI). The algorithm based on the parallel BQ decomposition used in the frame of the MPC algorithm for three-stage stochastic problems, was implemented in the Fortran 90 programming language and executed on cluster of SMPs. The performance results of the experiments are discussed in the chapter.