Cluster Of Workstations Research Articles

Adaptive data parallel computing on workstation clusters

Many important parallel applications are data parallel, and may be efficiently implemented on a workstation cluster by allocating each workstation a contiguous partition of the data domain. Implementation on non-dedicated clusters, however, is complicated by the possibility of changes in workstation availability. For example, a personal workstation may be reclaimed by its primary user for interactive use. In such situations, a node must be removed from the collection of workstations forming the “virtual parallel machine” allocated to the application, and data redistributed accordingly. Conversely, workstations may become available to join the virtual parallel machine. This paper identifies fundamental characteristics of efficient policies for data redistribution following addition/removal of workstations from the cluster. The following conclusions are obtained based on mathematical analysis and simulations: (a) allocating data to a new node from the center of the data domain substantially reduces data migration costs compared to allocation from the edge; (b) addition in groups is beneficial compared to repeated single additions; and (c) even a large number of incremental adjustments of the data domain partitions, owing to successive additions/removals of nodes, do not appear to substantially degrade partition quality compared to that obtained by partitioning from scratch. We believe that these observations can be fruitfully incorporated in the design of workstation cluster support systems for data parallel computing.

A networked client–server environment with CORBA interface for parallel FE analysis

The paper describes the development of a powerful high performance finite element computing framework in a network of small computers. A basic object-oriented general purpose FE analysis code with a frontal solver has been developed in C++. CORBA has been used as a platform for objects deployment as a network in a client–server environment. As mentioned earlier, the server has been developed in C++ while both Java and C++ clients have been used. Different cases viz. (i) simple client/server model and (ii) multiple concurrent servers with a Java client model have been developed and validated. The distributed network environment has been utilized for parallelizing the finite element analysis. This has presently been implemented over a cluster of Intel workstations connected through a switch, thereby offering a very economical solution for the solution of large problems. Sub-structuring and use of superelements have been employed for this parallelization. Each sub-server is responsible for the assembly of elements, computation of superelement stiffness, communication with central server and the computation of low-level solution. Some initial results on performance are also reported.

Parallel Line Search in Method of Feasible Directions

In this paper the line search procedure within the method of feasible directions is parallelized and used in the solution of constrained structural optimization problems. As the objective function is linear in the variables, the step size problem reduces to a zero finding problem. That is, the step size is the distance along the direction vector to the nearest constraint boundary. Zero finding is accomplished in two steps—a ‘march’ along the direction vector to bracket the zero followed by an interval reduction scheme. Both these steps are parallelized using MPI for message passing. When implemented on a cluster of workstations, for a convergence parameter of 10−6, the time for optimization of composite pressure vessel reduces from 3\(\frac{1}{2}\) hours to \(\frac{1}{2}\) hour when 64 processors are utilized, with a speedup of 7.0.

Cluster communication protocols for parallel-programming systems

Clusters of workstations are a popular platform for high-performance computing. For many parallel applications, efficient use of a fast interconnection network is essential for good performance. Several modern System Area Networks include programmable network interfaces that can be tailored to perform protocol tasks that otherwise would need to be done by the host processors. Finding the right trade-off between protocol processing at the host and the network interface is difficult in general. In this work, we systematically evaluate the performance of different implementations of a single, user-level communication interface. The implementations make different architectural assumptions about the reliability of the network and the capabilities of the network interface. The implementations differ accordingly in their division of protocol tasks between host software, network-interface firmware, and network hardware. Also, we investigate the effects of alternative data-transfer methods and multicast implementations, and we evaluate the influence of packet size. Using microbenchmarks, parallel-programming systems, and parallel applications, we assess the performance of the different implementations at multiple levels. We use two hardware platforms with different performance characteristics to validate our conclusions. We show how moving protocol tasks to a relatively slow network interface can yield both performance advantages and disadvantages, depending on specific characteristics of the application and the underlying parallel-programming system.

A hybrid self‐organizing maps and particle swarm optimization approach

AbstractGene clustering, the process of grouping related genes in the same cluster, is at the foundation of different genomic studies that aim at analyzing the function of genes. Microarray technologies have made it possible to measure gene expression levels for thousands of genes simultaneously. For knowledge to be extracted from the datasets generated by these technologies, the datasets have to be presented to a scientist in a meaningful way. Gene clustering methods serve this purpose. In this paper, a hybrid clustering approach that is based on self‐organizing maps and particle swarm optimization is proposed. In the proposed algorithm, the rate of convergence is improved by adding a conscience factor to the self‐organizing maps algorithm. The robustness of the result is measured by using a resampling technique. The algorithm is implemented on a cluster of workstations. Copyright © 2004 John Wiley & Sons, Ltd.

Local DNA sequence alignment in a cluster of workstations: Algorithms and tools

Abstract Distributed Shared Memory systems allow the use of the shared memory programming paradigm in distributed architectures where no physically shared memory exist. Scope consistent software DSMs provide a relaxed memory model that reduces the coherence overhead by ensuring consistency only at synchronization operations, on a per-lock basis. Much of the work in DSM systems is validated by benchmarks and there are only a few examples of real parallel applications running on DSM systems. Sequence comparison is a basic operation in DNA sequencing projects, and most of sequence comparison methods used are based on heuristics, that are faster but do not produce optimal alignments. Recently, many organisms had their DNA entirely sequenced, and this reality presents the need for comparing long DNA sequences, which is a challenging task due to its high demands for computational power and memory. In this article, we present and evaluate a parallelization strategy for implementing a sequence alignment algorithm for long sequences. This strategy was implemented in JIAJIA, a scope consistent software DSM system. Our results on an eight-machine cluster presented good speedups, showing that our parallelization strategy and programming support were appropriate.

A Study on Global and Local Optimization Techniques for TCAD Analysis Tasks

We evaluate optimization techniques to reduce the necessary user interaction for inverse modeling applications as they are used in the technology computer-aided design field. Four optimization strategies are compared. Two well-known global optimization methods, simulated annealing and genetic optimization, a local gradient-based optimization strategy, and a combination of a local and a global method. We rate the applicability of each method in terms of the minimal achievable target value for a given number of simulation runs and in terms of the fastest convergence. A brief overview over the three used optimization algorithms is given. The optimization framework that is used to distribute the workload over a cluster of workstations is described. The actual comparison is achieved by means of an inverse modeling application that is performed for various settings of the optimization algorithms. All presented optimization algorithms are capable of evaluating several targets in parallel. The best optimization strategy that is found is used in the calibration of a model for silicon self-interstitial cluster formation and dissolution.

Optimizing the execution of multiple data analysis queries on parallel and distributed environments

We investigate techniques for efficiently executing multiquery workloads from data and computation-intensive applications in parallel and/or distributed computing environments. In this context, we describe a database optimization framework that supports data and computation reuse, query scheduling, and active semantic caching to speed up the evaluation of multiquery workloads. Its most striking feature is the ability of optimizing the execution of queries in the presence of application-specific constructs by employing a customizable data and computation reuse model. Furthermore, we discuss how the proposed optimization model is flexible enough to work efficiently irrespective of the parallel/distributed environment underneath. In order to evaluate the proposed optimization techniques, we present experimental evidence using real data analysis applications. For this purpose, a common implementation for the queries under study was provided according to the database optimization framework and deployed on top of three distinct experimental configurations: a shared memory multiprocessor, a cluster of workstations, and a distributed computational Grid-like environment.

Speed‐up of radiation treatment planning using a workstation cluster

AbstractIn treatment planning in a heavy particle cancer radiation facility which is utilized by multiple remote medical institutions, it is necessary to calculate accurately the three‐dimensional dose distribution within the body of the patient in order to achieve efficient treatment. However, this presents a problem, since a long processing time is needed, which degrades the efficiency of treatment planning. In order to handle this problem, the authors used a cluster system in which Alpha 21164A CPUs were connected over a 100 Mbit/s Ethernet to speed up the dose distribution calculation (Dose) and digital reconstructed X‐ray image regeneration (DRR) by parallel processing. Parallel calculation is performed by segmenting the calculation area. The communication time among processors is reduced by data compression, and the load is made uniform by assigning more than one area to a processor. As a result of evaluation experiments, calculations performed iteratively by varying parameters at the actual site of medical care were accelerated by using 10 processors, by a factor of 7 for Dose, and 9 for DRR. In other words, the speed was improved in proportion to the number of processors. This implies that processing which has required several tens of seconds can now be handled in 3 to 6 seconds, which is short enough for the user to wait before the terminal. The objective is to make the system usable by multiple remote medical institutions by refining system functions such as fault recovery in order to ensure its reliability as a calculation server. © 2004 Wiley Periodicals, Inc. Syst Comp Jpn, 35(8): 96–106, 2004; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.10277

Using Distributed Computers to Deterministically Approximate Higher Dimensional Convection Diffusion Equations

A new approach to solving D > 3 spatial dimensional convection-diffusion equation on clusters of workstations is derived by exploiting the stability and scalability of the combination of a generalized D dimensional high-order compact (HOC) implicit finite difference scheme and parallelized GMRES(m). We then consider its application to multifactor Option pricing using the Black–Scholes equation and further show that an isotropic fourth order compact difference scheme is numerically stable and determine conditions under which its coefficient matrix is positive definite. The performance of GMRES(m) on distributed computers is limited by the inter-processor communication required by the matrix-vector multiplication. It is shown that the compact scheme requires approximately half the number of communications as a non-compact difference scheme of the same order of truncation error. As the dimensionality is increased, the ratio of computation that can be overlapped with communication also increases. CPU times and parallel efficiency graphs for single time step approximation of up to a 7D HOC scheme on 16 processors confirm the numerical stability constraint and demonstrate improved parallel scalability over non-compact difference schemes.

Modeling Clustered Task Graphs for Scheduling Large Parallel Programs in Distributed Systems

The recent development of distributed processing platforms, such as clusters of workstations, makes the use of distributed applications more extensive. A fundamental issue affecting performance in distributed systems is the scheduling of tasks to processors. The problem of solving the scheduling of parallel programs modelled with a task precedence graph (TPG) has been extensively studied. Scheduling algorithms of TPGs can be solved by a one-step method when a fixed number of processors is considered or through a two-step method by first creating an unbounded number of groups of tasks (clusters) and subsequently assigning these clusters to a bounded number of processors. The goal of this work is to present a new mapping algorithm called TASC (Task ASsignment exploiting Concurrency) to solve this second step of assigning clusters to processors. The effectiveness of TASC is established through simulation for a set of synthetic graphs that model real applications.

Inexact block Newton methods for solving nonlinear equations

In the paper two parallelizable inexact block Newton methods are presented for solving large and sparse nonlinear equations. The basic idea is simple and direct. Combining the simplified Newton method with the component averaging (CAV) method [Parallel Comput. 27 (2001) 777] results in an inexact Newton method, called simplified Newton-CAV method. Parallel tests of the algorithm are implemented on a message-passing distributed-memory multiprocessor architecture such as a cluster of workstations. The results show that the new algorithm can achieve good performance. Moreover as a development of the simplified Newton-CAV method, the overlapped block Newton-CAV method is further proposed and discussed.

Numerical Performance of the Distributed Vector Finite-Element Time-Domain Algorithm

This paper deals with a distributed time-domain modeling of electromagnetic phenomena with the finite-element method. The model is approximated by edge elements. The constitutive equations and method of parallelization of the algorithm are presented. The properties of the distributed finite-element time-domain algorithm are discussed. Some typical performance metrics are studied for the parallel versions of the software. The presented algorithm is executed on a heterogeneous and a homogeneous clusters of workstations. Two different distributed memory environments (MPI and PVM) are used to evaluate the efficiency of the algorithm.

Modeling the dewatering and depressurization of the Lihir open-pit gold mine, Papua New Guinea

The open pit at the Lihir Gold Mine, Papua New Guinea, is planned ultimately to reach more than 200 m below sea level. Cooling and depressurization of the geothermal resource associated with the gold mineralisation is an essential part of the mining operation. This paper deals with the development of a numerical model of the resource capable of representing reservoir processes associated with shallow dewatering, geothermal development and providing sufficient detail for pit-wall stability calculations. Use was made of iTOUGH2 [Finsterle, 1999. iTOUGH2 User’s Guide, Report LBNL-40040] running on a cluster of LINUX workstations to aid the fitting of some model parameters. Use of this program in a parallel computational environment was essential to complete the parameter fitting in an acceptable time. A detailed model containing almost 90,000 elements has been developed as an aid to pit-wall stability calculations and this has been run on a Cray and a LINUX cluster using a multiprocessor version of TOUGH2 recently developed at LBNL [ Zhang et al., 2001. Parallel computing technique for large-scale reservoir simulation of multi-component and multiple fluid flow].

The design of a distributed MATLAB-based environment for computing pseudospectra

It has been documented in the literature that the pseudospectrum of a matrix is a powerful concept that broadens our understanding of phenomena based on matrix computations. When the matrix A is non-normal, however, the computation of the pseudospectrum becomes a very expensive computational task. Thus, the use of high performance computing resources becomes key to obtaining useful answers in acceptable amounts of time. In this work we describe the design and implementation of an environment that integrates a suite of state-of-the-art algorithms running on a cluster of workstations to enable the matrix pseudospectrum become a practical tool for scientists and engineers. The user interacts with the environment via the graphical user interface PPsGUI. The environment is constructed on top of CMTM, an existing environment that enables distributed computation via an MPI API for MATLAB.

Efficient resource management applied to master–worker applications

We investigate the problem arising in scheduling parallel applications that follow a master–worker paradigm in order to maximize both resource efficiency and application performance. Based on the results obtained in a previous simulation study, we have derived a self-adjusting strategy that can be used to dynamically adjust the number of processors allocated to the application. The effectiveness of the proposed strategy has been assessed in two different scenarios: first, we implemented and tested this strategy on a cluster of homogeneous workstations. Secondly, we extended the self-adjusting strategy to be applied on heterogeneous clusters. We assessed the effectiveness of our strategy using an image-thinning application as a practical example of master–worker application.

The design and implementation of a modular and extensible Java Virtual Machine

AbstractThis paper describes the design, implementation, and experimental evaluation of a modular and extensible Java™ Virtual Machine (JVM) infrastructure, called Jupiter. The infrastructure is intended to serve as a vehicle for our research on scalable JVM architectures for a cluster of PC workstations, with support for shared memory in software. Jupiter is constructed, using a building block architecture, out of many modules with small, simple interfaces. This flexible structure, similar to UNIX® shells that build complex command pipelines out of discrete programs, allows the rapid prototyping of our research ideas by confining changes in JVM design to a small number of modules. In spite of this flexibility, Jupiter delivers good performance. Experimental evaluation of the current implementation of Jupiter using the SPECjvm98 and the EPCC Java Grande single‐threaded and multithreaded benchmarks reflects competitive performance. Jupiter is on average about 2.5 times faster than Kaffe and about 2 times slower than the Sun Microsystems JDK (interpreter versions only). By providing a flexible JVM infrastructure that delivers competitive performance, we believe we have developed a framework that supports further research into JVM scalability. Copyright © 2003 John Wiley & Sons, Ltd.

Evaluating linear recursive filters on clusters of workstations

The aim of this paper is to show that the recently developed high performance algorithm for solving linear recurrence systems with constant coefficients together with the new BLAS-based algorithm for narrow-banded triangular Toeplitz matrix-vector multiplication allow to evaluate linear recursive filters efficiently, even on clusters of workstations. The results of experiments performed on a cluster of twelve Linux workstations are also presented. The performance of the algorithm is comparable with the performance of two processors of Cray SV-1 for such kind of recursive problems.

Quick Matrix Multiplication on Clusters of Workstations

A quick matrix multiplication algorithm is presented and evaluated on a cluster of networked workstations consisting of Pentium hosts connected together by Ethernet segments. The obtained results confirm the feasibility of using networked workstations to provide fast and low cost solutions to many computationally intensive applications such as large linear algebraic systems. The paper also presents and verifies an accurate timing model to predict the performance of the proposed algorithm on arbitrary clusters of workstations. Through this model the viability of the proposed algorithm can be revealed without the extra effort that would be needed to carry out real testing.

Optimization of Flapping Airfoils for Maximum Thrust and Propulsive Efficiency

A numerical optimization algorithm based on the steepest decent along the variation of the optimization function is implemented for maximizing the thrust and/or propulsive efficiency of a single flapping airfoil. Unsteady, low speed laminar flows are computed using a Navier-Stokes solver on moving overset grids. The flapping motion of the airfoil is described by a combined sinusoidal plunge and pitching motion. Optimization parameters are taken to be the amplitudes of the plunge and pitching motions, and the phase shift between them. Computations are performed in parallel in a work station cluster. The numerical simulations show that high thrust values may be obtained at the expense of reduced efficiency. For high efficiency in thrust generation, the induced angle of attack of the airfoil is reduced and large scale vortex formations at the leading edge are prevented.