Markus Schwehm

Next century challenges: Nexus—an open global infrastructure for spatial-aware applications

Due to the lack of a generic platform for locationand spatial-aware systems, many basic services have to be reimplemented in each application that uses spatial-awareness. A cooperation among different applications is also difficult to achieve without a common platform. In this paper we present a platform that solves these problems. It provides an infrastructure that is based on digital models of regions of the physical world, which are augmented by virtual objects. We show how virtual objects make the integration of existing information systems and services in spatial-aware systems easier. Furthermore, our platform supports interactions between the computer models and the real world and integrates single models in a global “Augmented World”.

Mole 3.0: a middleware for Java-based mobile software agents

Mole is one of the first Java-based mobile agent systems. It runs on an unmodified Java virtual machine on PCs under Windows 95/NT and on workstations under several UNIX dialects. Earlier versions of Mole have provided a basic infrastructure for communication and migration of mobile agents. Version 3.0 of Mole has been strongly revised and several requests and proposals from users of the earlier versions of Mole were integrated into the new release. In particular Mole supports communication between agent groups and the concept of sessions. The infrastructure of Mole includes a resource manager, a directory service and a global naming scheme for agents. In order to support the design of agents, a graphical agent monitor allows to visualize the system behavior as a whole or of a single agent in particular. Mole further provides a thread management unit to overcome some shortcomings of the Java virtual machine. Mole provides a simple means for installation and configuration of the system. This paper describes the architecture and infrastructure of Mole 3.0.

Mapping and Scheduling by Genetic Algorithms

A massively parallel genetic algorithm for the mapping and scheduling problem is presented. It turns out that a standard genetic algorithm package can easily be adapted to the mapping and scheduling problem. The resulting algorithm is able to exploit parallelism of a massively parallel hardware and can solve larger problems than a reference algorithm. Moreover the algorithm is well suited if the algorithm has to be restarted with a slightly modified problem input. The algorithm is implemented on the array processor MasPar MP-1.

A Massively Parallel Genetic Algorithm on the MasPar MP-1

This contribution describes the implementation of a fine-grained parallel genetic algorithm ‘MPGA’ on the MasPar MP-1, a massively parallel mesh connected array processor with global router and 1024 (up to 16384) 4-bit processing elements. The implementation uses object oriented methods to provide a large set of standard strategies which can be adapted for a given application. Report modules support the investigation of the performance of the GA. The Implementation shows a good performance compared to other implementations on parallel hardware.

Determining the Optimal Network Partition and Kanban Allocation in JIT Production Lines

One way to reduce costs in high volume production lines is to smooth and balance the material flow by means of controlled inventories. Kanban systems are now being implemented worldwide due to their ability of reducing inventories and production lead times. This paper addresses two fundamental design issues in kanban systems and presents an efficient heuristic method for designing such systems. An analytical method for modelling kanban systems and a general-purpose genetic algorithm are integrated in a heuristic design methodology which evaluates the performance of kanban systems using alternative network partitions and allocations of kanbans. As we demonstrate, the heuristic method provides a useful procedure to evaluate the impact of design alternatives and can thus serve as a rough-cut decision support tool which assists managers in the planning of largescale manufacturing systems.

Scheduling of parallel programs on configurable multiprocessors by genetic algorithms

Abstract The scheduling of programs on parallel hardware is investigated in order to minimize the response time of the resulting system. In particular the scheduling problem considered in this paper includes — next to the search for an optimal mapping of the tasks and their sequence of execution — also the search for an optimal configuration of the parallel hardware. An approach for the simultaneous optimization of all three components using genetic algorithms is presented and its performance is evaluated in comparison with an exact reference method based on an branch-and-bound-with-underestimates algorithm. The comparison is based on a large set of problem instances and includes regular task graphs with varying structure and scalable size, which had to be mapped onto a configurable parallel hardware consisting of 4–16 transputers, respectively. Small problem instances with less than eight tasks can be solved by both solution methods. For larger problem instances the reference method fails due to runtime complexity while the genetic algorithm can still find (suboptimal) solutions for problem instances with up to 120 tasks in an acceptable amount of time.

Massively Parallel Genetic Algorithms.

The genetic algorithm is an iterative random search technique for nonlinear or combinatorial problems. In this contribution, first the development from the classical genetic algorithm (GA) via the parallel genetic algorithm (PGA) to the massively parallel genetic algorithm (MPGA) is described. Then experimental results with an implementation of the MPGA on the array processor MasPar MP-1 are displayed, which exemplify robustness and adaptive behavior of the algorithm. The observed properties of the MPGA are finally combined for an improved method of mapping load onto a massively parallel hardware.

Comparing the DAP, Meiko and Suprenum with a Fluid Dynamic Benchmark

An algorithm for the numerical simulation of the fluid flow in a crystal growth process is presented. The algorithm is implemented on three parallel architectures. The performance analysis shows that special care has to be taken for the efficient realization of communication patterns on massively parallel systems.