Steffen Limmer

Performance investigations of genetic algorithms on graphics cards

Abstract Genetic algorithms are one of the most adaptable optimization algorithms. Due to their inherent parallelism they seem well suited for the execution on massively parallel hardware such as graphics processing units. In this paper we put this claim to the test by performing comprehensive experiments. We try to find out how well graphics processing units are suited for the task and what parts of genetic algorithms should be executed on them. We focus especially on the new Fermi generation of Nvidia graphics chips. While it is imperative the fitness function be effectively parallelizable on the GPU, because it is the most computational expensive task of the algorithm, results indicate that if this is the case, speedups of several orders of magnitude are possible compared to conventional multi-core CPUs. Our findings also suggest that, starting with the Fermi architecture, all parts of a genetic algorithm should be carried out on the graphics card instead of only part of it.

Accuracy and performance of 3D mask models in optical projection lithography

Different mask models have been compared: rigorous electromagnetic field (EMF) modeling, rigorous EMF modeling with decomposition techniques and the thin mask approach (Kirchhoff approach) to simulate optical diffraction from different mask patterns in projection systems for lithography. In addition, each rigorous model was tested for two different formulations for partially coherent imaging: The Hopkins assumption and rigorous simulation of mask diffraction orders for multiple illumination angles. The aim of this work is to closely approximate results of the rigorous EMF method by the thin mask model enhanced with pupil filtering techniques. The validity of this approach for different feature sizes, shapes and illumination conditions is investigated.

Framework for distributed evolutionary algorithms in computational grids

In the recent years an increasing number of computational grids have been built, providing an unprecedented amount of computational power. Based on their inherent parallelism, Evolutionary Algorithms are well suited for distributed execution in such grids. Unfortunately, there are several challenges concerning the usage of a grid infrastructure (e.g. the synchronization and submission of jobs and file transfer tasks). In this paper we present a new framework which makes a Globus based grid easily accessible for Evolutionary Algorithms and takes care of the parallelization. The usability is demonstrated by the example of an Evolutionary Algorithm for the Traveling Salesman Problem.

Comparison of common parallel architectures for the execution of the island model and the global parallelization of evolutionary algorithms

Evolutionary algorithms are one of the most popular forms of optimization algorithms. They are comparatively easy to use and were successfully employed for a wide variety of practical applications. However, frequently, it is necessary to execute them in parallel in order to reduce the runtime. There are a number of different approaches for the parallelization of evolutionary algorithms, and various hardware platforms can be used for the parallel execution. However, not every platform is equally suitable for any kind of parallelization of evolutionary algorithms. In addition, it also depends on properties of the concrete optimization problem to be solved and on the used evolutionary algorithm, which platform is best suited for the execution. The present work observes this in detail for two common forms of parallelization of evolutionary algorithms – the island model and the global parallelization – and for four widely used parallel computing platforms – multi‐core CPUs, clusters, graphics cards, and grids. Based on empirical and analytical investigations, it is determined, under which circumstances an architecture is better suited for the execution of a parallel evolutionary algorithm than another (and vice versa). Guidelines are derived that support users of parallel evolutionary algorithms with the choice of an appropriate platform. Copyright

Evolutionary optimization of layouts for high density free space optical network links

Electrical chip- and board-level connections are becoming more and more a bottleneck in computation. A solution to that problem could be optical connections, which allow a higher bandwidth. The usage of free space optics can avoid the problem of crosstalk and geometrical signal path crossings in systems with a high density of interconnections. The choice of appropriate design parameters, allowing the realization of such interconnections, is a complicated task. We present an evolutionary algorithm that is able to find these parameters. We describe the parallel execution of that algorithm and present optimization results.

Performance Investigation and Tuning in the Interoperable Cloud4E Platform

The paradigm of Software as a Service (SaaS) offers an interesting option to vendors of simulation software for providing their applications to a wide circle of customers. However, this imposes a challenge to vendors whose applications exist only as classical desktop tools, so far: Interfaces for the remote control have to be provided which are as independent from the underlying cloud infrastructure as possible in order to avoid vendor lock-ins. We present an interoperable platform developed in the project Cloud4E (Trusted Cloud Computing for Engineering), which allows the provisioning of existing simulation software in form of a service in a cloud. The interoperability of the platform and of the services is achieved by the usage of the Open Cloud Computing Interface (OCCI) together with the Advanced Message Queuing Protocol (AMQP) where OCCI is not only used as interface to Infrastructure as a Service (IaaS) but also as interface to SaaS. Hence, the OCCI server plays a central role within the platform and can quickly become a bottleneck, which degrades the performance of the whole platform. We present detailed performance investigations and suggest options to improve the performance. The investigations were performed on the widely used OCCI server implementation called rOCCI server connected to the OpenNebula cloud middleware.

Services for numerical simulations and optimisations in grids

Numerical simulations are typically very compute intensive. A way to satisfy the arising demands for compute power is the employment of grid technology. But today, there exist only few tools, supporting the user in using grids for numerical simulations. Additionally, some basic problems of grid technology, are not solved with satisfaction. The goal of the project OptiNum-Grid was, to address these issues. We present a number of tools that were developed in the project. These are tools for parameter sweeps and optimizations in grids, for data security and automatic software installation, as well as for virtual environments in grids.

Three-dimensional crossbar interconnection using planar-integrated free-space optics and digital mirror-device

We consider the implementation of a dynamic crossbar interconnect using planar-integrated free-space optics (PIFSO) and a digital mirror-device™ (DMD). Because of the 3D nature of free-space optics, this approach is able to solve geometrical problems with crossings of the signal paths that occur in waveguide optical and electrical interconnection, especially for large number of connections. The DMD device allows one to route the signals dynamically. Due to the large number of individual mirror elements in the DMD, different optical path configurations are possible, thus offering the chance for optimizing the network configuration. The optimization is achieved by using an evolutionary algorithm for finding best values for a skewless parallel interconnection. Here, we present results and experimental examples for the use of the PIFSO/DMD-setup.

Investigation of strategies for an increasing population size in multi-objective CMA-ES

The Multi-objective Covariance Matrix Adaptation Evolution Strategy (MO-CMA-ES) is an evolutionary algorithm for continuous vector optimization. It is invariant against rotations and translations of the search space and empirical evaluations have shown that it is very competitive with other popular multi-objective evolutionary algorithms, like NSGA-II. However, MO-CMA-ES requires a certain “warm-up phase” to adapt internal strategy parameters. A promising approach to speed up this “warm-up phase” is to keep the start population small and to gradually increase its size during the optimization. We experimentally investigate two static and three dynamic strategies for increasing the population size. The results show that the employment of a dynamic population size increasing strategy can significantly improve the performance of MO-CMA-ES, especially when the budget of objective function evaluations is small.

Porting of the transfer-matrix method for multilayer thin-film computations on graphics processing units

Abstract. Thin-film computations are often a time-consuming task during optical design. An efficient way to accelerate these computations with the help of graphics processing units (GPUs) is described. It turned out that significant speed-ups can be achieved. We investigate the circumstances under which the best speed-up values can be expected. Therefore we compare different GPUs among themselves and with a modern CPU. Furthermore, the effect of thickness modulation on the speed-up and the runtime behavior depending on the input data is examined.