PC-based Cluster Research Articles

Ad-hoc cluster and workflow for parallel implementation of initial-stage evolutionary optimum design

Numerical optimization and especially topology and shape optimization are typically numerically very comprehensive due to implicit models, large number of variables, high non-linearity, possible infeasibility of candidate designs and related causes. It usually has to be implemented by coupling heterogeneous program environments such as evolutionary optimizers, computer-aided geometric modeling programs and simulators such as finite-element based analysis packages. This paper develops a workflow-based numerical system that serves the purpose of integrating, harmonizing and managing these distinct components in such a way that it provides the functionality needed for the inverse problem of evolutionary initial-stage topology- and shape synthesis. The approach based on inserting and optimizing holes is developed using chained piecewise Bezier curves and/or surfaces for shape parameterization. The procedure developed here employs existing off-the-shelf software packages for computer-aided design, finite element analysis and numerical optimization while building custom middle-ware programs and scripts. It also provides for parallel invoking of different and/or multiple simulators which reduces the optimization run-time by widening the critical bottle-necks in the overall process. The system is implemented inexpensively as an ad-hoc PC-based cluster where individual computers expose server programs and respective services which control locally installed simulators.

The design and implementation of Visuel performance monitoring and analysis toolkit for cluster and grid environments

The computing power provided by high performance and low-cost PC-based clusters with Grid platforms are attractive and they are equal or superior to supercomputers and mainframes. In this paper, we present implementation and design rationale of Visuel toolkit for MPI parallel program performance measurement and analysis in cluster and grid environments. Most of performance visualization tools available today for high-performance platforms show solely system performance data (e.g., CPU load, memory usage, network bandwidth, server average load), and thus, being suitable for computing system activity visualization. The Visuel (Visuel (in French language) = to visualize) toolkit is web-based interface designed to show performance activities of all computing nodes of a distributed environment involved in the execution of MPI parallel program, such as CPU load level and memory usage of each computing node. In addition, this toolkit is able to display comparative performance data charts of MPI parallel applications and multiple executions under investigation. The usage of this toolkit shows that it outperforms in easing the process of investigation of parallel applications.

On development of an efficient parallel loop self-scheduling for grid computing environments

The approaches to deal with scheduling and load balancing on PC-based cluster systems are famous and well-known. Self-scheduling schemes, which are suitable for parallel loops with independent iterations on cluster computer system, they have been designed in the past. In this paper, we propose a new scheme that can adjust the scheduling parameter dynamically on an extremely heterogeneous PC-based cluster and Grid computing environments in order to improve system performance. A Grid computing environment consists of multiple PC-based clusters is constructed using Globus Toolkit and MPICH-G2 middleware. The experimental results show that our scheduling can result in higher performance than other similar schemes on Grid computing environments.

A light-weight, collaborative temporary file system for clustered Web servers

Previous studies indicate that I/O could become a performance bottleneck in commodity PC-based cluster Web servers. Current local native file systems do not work well for expensive file I/Os while specialized file systems have a limitation on portability. In this paper, we present a lightweight, collaborative temporary file system (CTFS) to improve disk I/O performance for clustered Web servers. CTFS employs several techniques to achieve high-performance, good scalability and portability: (1) a lightweight local temporal file system at each node, (2) using Remote Direct Memory Access (RDMA) to improve intra-cluster communication performance, and (3) a location-aware summary cache for scalable file-to-server lookup. Comprehensive trace-driven simulation experiments conclude that CTFS achieves up to a 37% better system throughput and reduces up to 47% total disk I/O latency than a local asynchronous FFS solution.

An Enhanced Parallel Loop Self-Scheduling Scheme for Cluster Environments

Approaches for dealing with scheduling and load-balancing in PC-based cluster systems are famous and well known. In such environments, Self-Scheduling Schemes are suitable for parallel loops with independent iterations. However, while schemes such as FSS, GSS, and TSS fit most computer systems, they cannot provide good load-balancing. Chao-Tung Yang and Shun-Chi Chang proposed a parallel loop scheduling scheme for heterogeneous PC cluster systems in Yang and Chang [13]. Though the proposed scheme allows users to choose parameters before execution initialization, weaknesses in it motivated us to develop further improvements. For instance, using fixed and monotonous parameters can easily lead to invalid scheduling due to use of previously input information. Thus, in this paper we propose a new scheme that fits most widely available computer systems and allows the scheduling parameter to be adjusted dynamically in order to provide higher overall performance.

A simple photochemical module implemented in RAMS for tropospheric ozone concentration forecast in the metropolitan area of São Paulo, Brazil: Coupling and validation

An in-line simplified photochemical module was coupled with RAMS, an atmopsheric mesoscale model, to produce an operational forecast of ozone and other constituent concentration for the Metropolitan Area of São Paulo (MASP), Brazil. Comparisons between surface pollutants concentrations data, provided by automatic network stations for air quality monitoring and modeled values, show good agreement, with correlation coefficients greater than 0.7 for all stations considered. Mean values over all analyzed stations have an index of agreement higher than 0.8. An important feature observed during the simulations was the ability of the model to simulate nocturnal concentrations of ozone over MASP which are usually observed as a result of transport of this pollutant from other regions and from higher levels of the atmosphere, representing the close coupling between the micro scale process and the chemical reactions, although these nocturnal maximums are overestimated by the model in some cases. The computational efficiency of the model shows that it can be considered for operational procedures with relatively small PC-based clusters and used as a tool for emission control strategies by environmental agencies concerned with the improvement of air quality.

A306 汎用GPUを用いた大規模可視化クラスタの構築

This paper reports the Visualization Subsystem for the Sensable Simulation System(Scube) which the authors have been developing. Scube is a 64-nodes PC-based cluster system in which a commodity GPU as the visualization accelerator is configured with each node. There is no dedicated special purpose network for the visualization, however, the high-cost performance inter-connection network is originally designed for Scube. All the hardware components for this network is essentially the small-scale and commodity hardware designed for Giga-bit Ethernet. The authors have also designed the binary-tree based parallel image composition algorithm which utilizes the degree of overlapping between sub-images, in order to reduce the amount of unnecessary communication and computation. The details of the image composition algorithm and the network configuration are described in this paper.

Parallelization of EMAP3D based on element-by-element Jacobi preconditioned conjugate gradient method

The demand for more accurate analyses of electromechanical systems, such as electromagnetic launchers and pulsed rotating power supplies, requires an increase in the size of the finite-element model of these systems. It is not uncommon for the number of unknowns for such a model to reach a half million. A parallel computing system with multiple processors and distributed memory, such as a PC cluster, makes it possible to obtain solutions for large-scale problems in reasonable times. In order to utilize this parallel hardware architecture, the software needs to be parallelized accordingly. Electromechanical Analysis Program in Three Dimensions (EMAP3D) is parallelized based on the element-by-element Jacobi preconditioned conjugate gradient (EBEJPCG) method because it is easily adopted into a parallel scheme and has low memory requirements because the formation of the global matrix is not necessary. The details of this algorithm are described in this paper. A block armature railgun was used to investigate this parallel algorithm on the Institute for Advanced Technology's (IAT's) eight-node PC-based Beowulf cluster. The performance of the algorithm in terms of speed-up ratio is presented.

Finite-element analyses of problems in electromagnetic launch using a PC-based Beowulf cluster with high-speed networks

Finite-element (FE) analyses in electromagnetic launch usually involve coupled electromagnetic, thermal, and structural equations. The number of unknowns to be solved could range up to a million and beyond. Parallel solution using massively parallel processing systems (MPPs) is an obvious approach toward reduction of computation times. Beowulf clusters are inexpensive parallel platforms compared with the supercomputing MPPs such as the Cray T3E. They could also serve as dedicated systems. The results from parametric analyses of a sliding electrical contact problem using an FE software EMAP3D and a Beowulf cluster are reported here.

Parallel Visualization of Large-scale Finite Element Analysis Result in Immersive Virtual Reality Space.

In large scale linite element (FE) analyses of over ten millions degrees of freedom (DOFs), it is very difficult not only to calculate but also to visualize them. This difficulty in visualization is mainly caused due to the following two reasons. Firstly, verification for large-scale analysis requires highly Interactive user interface. Secondary, it can not be handled on a conventional single processor machine any more. This paper describes a prototype of the visualization system for large scale FE solutions with millions of DOFs using a virtual reality environment named CABIN for the former issue and a PC-based cluster consisting of 24 PEs (DEC Alpha 533 MHz) for the latter issue. CABIN provides a large scale and realistic space in immersive multiscreen display. This makes it easy to understand the results, but requires excessively computer resources. Thus we employ the PC-based cluster, which supplies the processed data to the front-ended CABIN system. Practical performances of the present system are demonstrated for both three-dimensional elastic and fluid flow analyses.