Cluster Of Workstations Research Articles

Incremental DataGrid Mining Algorithm for Mobility Prediction of Mobile Users

Problem statement: Mobility prediction is the important issue in Personal Communication Systems (PCS). Mobile users moving logs are stored in data grid located in different locations. Distributed data mining algorithm is applied on this moving logs to generate the mobility pattern of mobile users. As new moving logs are added to the data grid, existing mobility pattern becomes invalid and it should be updated. One of the existing work to derive the new mobility pattern is re-executing the algorithm from scratch results in excessive computation. Approach: We had designed new incremental algorithm by maintaining infrequent mobility patterns, which avoids unnecessary scan of full database. Incremental data mining algorithm taken lesser time to compute new mobility patterns. The discovered location patterns can be used to provide various location based services to the mobile user by the application server in mobile computing environment. Data grid provided geographically distributed database for computational grid which implements incremental data mining algorithm. We built data grid system on a cluster of workstation using open source globus toolkit 4.0 and Message Passing Interface extended with Grid Services (MPICH-G2). Results: The experiments were conducted on original data sets and data were added incrementally and the computation time was recorded for each data sets. The performance improvement for increment size of 100 K was about 55% for 0.20% support count and it is increased to 60% for 0.25% support count. The performance is increased about 65% for the support count 0.30%. Conclusion: We analyzed our results with various sizes of data sets and the proof shows the time taken to generate mobility pattern by incremental mining algorithm is less than re-computing approach. In future the execution time can further be reduced by balancing the workload of grid nodes.

Self-tuning management of update-intensive multidimensional data in clusters of workstations

Contemporary applications continuously modify large volumes of multidimensional data that must be accessed efficiently and, more importantly, must be updated in a timely manner. Single-server storage approaches are insufficient when managing such volumes of data, while the high frequency of data modification render classical indexing methods inefficient. To address these two problems we introduce a distributed storage manager for multidimensional data based on a Cluster-of-Workstations. The manager addresses the above challenges through a set of mechanisms that, through selective on-line data reorganization, collectively maintain a balanced load across a cluster of workstations. With the help of both a highly efficient and speedy self-tuning mechanism, based on a new data structure called stat-index, as well as a query aggregation and clustering algorithm, our storage manager attains short query response times even in the presence of massive modifications and highly skewed access patterns. Furthermore, we provide a data migration cost model used to determine the best data redistribution strategy. Through extensive experimentation with our prototype, we establish that our storage manager can sustain significant update rates with minimal overhead.

Radar image processing with clusters of computers

Some radar image processing algorithms such as shape-from-shading are particularly compute-intensive and time-consuming. If, in addition, a data set to be processed is large, then it may make sense to perform the processing of images on multiple workstations or parallel processing systems. We have implemented shape-from-shading, stereo matching, resampling, gridding, and visualization of terrain models in such a manner that they execute either on parallel machines or on clusters of workstations. We were motivated by the large image data set from NASA's Magellan mission to planet Venus, but received additional inspiration from the European Union's Center for Earth Observation program (CEO) and Austria's MISSION initiative for distributed processing of remote sensing images on remote workstations, using publicly-accessible algorithms. We developed a multi-processor approach that we denote as CDIP for Concurrent and Distributed Image Processing. The speedup for image processing tasks increases nearly linearly with the number of processors, be they on a parallel machine or arranged in a cluster of distributed workstations. Our approach adds benefits for users of complex image processing algorithms: the efforts for code porting and code maintenance are reduced and the necessity for specialized parallel processing hardware is eliminated.

Porting the OPATM-BFM Application to a Grid e-Infrastructure – Optimization of Communication and I/O Patterns

OPATM-BFM is an off-line three-dimensional coupled eco-hydrodynamic simulation model used for biogeochemical and ecosystem-level predictions. This paper presents the first results of research activities devoted to the adaptation of the parallel OPATM- BFM application for an efficient usage in modern Grid-based e-Infrastructures. For the application performance on standard Grid archi- tectures providing generic clusters of workstations, such results are important. We propose a message-passing analysis technique for communication-intensive parallel applications based on a preliminary application run analysis. This technique was successfully used for the OPATM-BFM application and allowed us to identify several optimization proposals for the current realization of the communication pattern. As the suggested improvements are quite generic, they can be potentially useful for other parallel scientific applications.

Broadcasting on Networks of Workstations

Broadcasting and multicasting are fundamental operations. In this work we develop algorithms for performing broadcast and multicast in clusters of workstations. In this model, sending a message to a machine in the same cluster takes 1 time unit, and sending a message to a machine in a different cluster takes C(≥1) time units. The clusters may have arbitrary sizes. Lowekamp and Beguelin proposed heuristics for this model, but their algorithms may produce broadcast times that are arbitrarily worse than optimal. We develop the first constant factor approximation algorithms for this model. Algorithm LCF (Largest Cluster First) for the basic model is simple, efficient and has a worst case approximation guarantee of 2. We then extend these models to more complex models where we remove the assumption that an unbounded amount of communication may happen using the global network. The algorithms for these models build on the LCF method developed for the basic problem. Finally, we develop broadcasting algorithms for the postal model where the sending node does not block for C time units when the message is in transit.

A Distributed Heuristic Algorithm for the Rectilinear Steiner Minimal Tree Problem

Rectilinear Steiner minimal tree (RSMT) problem finds a minimum length tree that interconnects a given set of points by only horizontal and vertical line segments and by using extra points if necessary. In this paper, to speedup the RSMT construction, two recently developed successful heuristic algorithms, namely rectilinear steiner tree (RST) by Zhou and batched greedy algorithm (BGA) by Kahng , have been used as the basis. Following a slight modification on RST, which led to a nonrecursive and a considerably faster version, a partially parallelized and distributed form of this modified algorithm is proposed. Computational tests using large random data sets have shown the advantage of the modification on RST, and tests conducted on a cluster of workstations have proven the proposed distributed approach to be very promising particularly for large problem instances.

Bandwidth optimal all-reduce algorithms for clusters of workstations

We consider an efficient realization of the all-reduce operation with large data sizes in cluster environments, under the assumption that the reduce operator is associative and commutative. We derive a tight lower bound of the amount of data that must be communicated in order to complete this operation and propose a ring-based algorithm that only requires tree connectivity to achieve bandwidth optimality. Unlike the widely used butterfly-like all-reduce algorithm that incurs network contention in SMP/multi-core clusters, the proposed algorithm can achieve contention-free communication in almost all contemporary clusters, including SMP/multi-core clusters and Ethernet switched clusters with multiple switches. We demonstrate that the proposed algorithm is more efficient than other algorithms on clusters with different nodal architectures and networking technologies when the data size is sufficiently large.

A COMPARATIVE STUDY OF PARALLELIZATION STRATEGIES FOR FRACTAL IMAGE COMPRESSION ON A CLUSTER OF WORKSTATIONS

In this paper we implement and compare the performance of the Message Passing Interface (MPI) static master-worker and three strategies of MPI task farm implementations for fractal image compression on a Beowulf cluster of workstations, namely Local Predecimation with Range Index Communication (LPRI), Global Predecimation with Range Communication (GPR) and No Predecimation with Range Index Communication (NPRI). Our results show that the MPI task farm implementations balance the load effectively among workers as compared to the MPI static master-worker implementation. The task farm strategies are compared by measuring their speedup and worker idle time cost.

Error handling techniques of genetic algorithms in parallel computing environment

It is easy to create parallel genetic algorithm software with master-slave type paralelization on a cluster of workstations. In a real situation the probability of errors in communication or in some of the slave processes during a long calculation is significant.In this paper we deal with different error handling strategies in master-slave type paralelization of standard GA algorithms and show results of test calculations. Our simulations are close to real applications in the sense that we examine the best achieved objective function value at a fixed wall clock time with different error handling strategies depending on the probability of errors and number of processors. Using these results we make suggestions on the selection of a good error handling method in different optimization problems.

An agent architecture for managing data resources in a grid environment

The agent paradigm has been successfully used in a large number of research areas. MAPFS, a parallel file system, constitutes one successful application of agents to the I/O field, providing a multiagent I/O architecture. The use of a multiagent system implies coordination and cooperation among its agents. MAPFS is oriented to clusters of workstations, where agents are applied in order to provide features such as caching or prefetching. The adaptation of MAPFS to a grid environment is named MAPFS-Grid. Agents can help to increase the performance of data-intensive applications running on top of the grid. This paper describes the conceptual agent framework and the communication model used in MAPFS-Grid, which provides the management of data resources in a grid environment. The evaluation of our proposal shows the advantages of using agents in a data grid.

Cluster computing of mechanisms dynamics using recursive formulation

This paper presents the O(n) recursive algorithm for forward dynamics of closed loop kinematic chains adapted to parallel computations on a cluster of workstations. The Newton–Euler equations of motion are formulated in terms of relative coordinates. Closed loop kinematic chains are transformed into open loop chains by cut joint technique. Cut joint constraint and Lagrange multipliers are introduced to complete the equations of motion. Constraint stabilization is performed using the Baumgarte stabilization technique with application to multibody systems with large number of degrees of freedom. Numerical simulations are carried out to study the influence of the degrees of freedom of the multibody system on computational efficiency of the algorithm using the Message Passing Interface (MPI). We also consider the ways of minimization of communication overhead which has significant impact on efficiency in case of cluster computing.

Natural convection flow in the cavity with isoflux boundaries

Natural convection flow in rectangular cavities with uniform heat flux side walls and an adiabatic floor and ceiling is investigated. The analytical solution for the evenly heated and cooled infinitely tall cavity, obtained by integrating the energy equation over a certain control volume, is introduced and compared to a full numerical solution for the finite cavity. Numerical solutions have been obtained for cavities with height-to-width ratios of 1 to 10, various values of the heat flux and with parameters appropriate to both air and water. For high enough aspect ratio and/or Rayleigh number the numerical solutions at mid-height of the cavity and the cavity stratification are well predicted by the analytical solution for the infinite cavity. The flow on the cavity side walls is then one dimensional. References Armfeld, S. W., and Patterson, J. C., Direct simulation of wave interactions in unsteady natural convection in a cavity. Int. J. Heat Mass Transfer 34 (1991) 929--940. doi:10.1016/0017-9310(91)90004-X Armfeld, S. W., and Patterson, J. C., Wave properties of natural convection boundary layers. J. Fluid Mech. 239 (1992) 195--211. Desrayaud, G., and Nguyen, T. H., Instabilities themoconvectives dans une cavite a flux imposes, in Douzieme Congres Canadien de Mecanique Appliquee (1989) 716--717. Jiracheewanun, S., Armfield, S. W., Behnia, M., and McBain, G. D., The transient behaviour of a differentially heated cavity with isoflux boundaries, Proc. 15th A'asian Fluid Mechanics Conf. (2004) Paper AFMC00246. Kimura, S. and Bejan, A., The boundary layer natural convection regime in a rectangular cavity with uniform heat flux from the side, J. Heat Transfer 106 (No. 1) (1984) 98--103. Leonard, B. P., A stable and accurate convective modeling procedure based on quadratic upstream interpolation, Comp. methods appl. mech. engr. 19 (1979) 59--98. doi:10.1016/0045-7825(79)90034-3 Lietzke, A. F., Theoretical and experimental investigation of heat transfer by laminar natural convection between parallel plates, (NACA, 1955) Report 1223. McBain, G. D., Armfield, S. W. and Jiracheewanun, S., The conduction and convection regimes in a cavity with evenly heated and cooled vertical walls, 8th Australasian Heat and Mass Transfer Conf. (2005). (in press) Norris, S. E. and Armfield, S. W., Solving the Navier--Stokes equations on a workstation cluster, ANZIAM J. 42(E) (2000) C1058--C1075. http://anziamj.austms.org.au/V42/CTAC99/Norr Patterson, J. C. and Armfield, S. W., Transient features of natural convection in a cavity, J. Fluid Mech. 219 (1990) 469--497. Patterson, J. and Imberger, J., Unsteady natural convection in a rectangular cavity, J. Fluid Mech. 100 (1980) 65--86.

Securing skeletal systems with limited performance penalty: The muskel experience

Algorithmic skeletons have been exploited to implement several parallel programming environments, targeting workstation clusters as well as workstation networks and computational grids. When targeting non-dedicated clusters, workstation networks and grids, security has to be taken adequately into account in order to guarantee both code and data confidentiality and integrity. However, introducing security is usually an expensive activity, both in terms of the effort required to managed security mechanisms and in terms of the time spent performing security related activities at run time. We discuss the cost of security introduction as well as how some features typical of skeleton technology can be exploited to improve the efficiency code and data securing in a typical skeleton based parallel programming environment and we evaluate the performance cost of security mechanisms implemented exploiting state of the art tools. In particular, we take into account the cost of security introduction in muskel, a Java based skeletal system exploiting macro data flow implementation technology. We consider the adoption of mechanisms that allow securing all the communications involving remote, unreliable nodes and we evaluate the cost of such mechanisms. Also, we consider the implications on the computational grains needed to scale secure and insecure skeletal computations.

Parallel implementation of the accelerated BEM approach for EMSI of the human brain

Boundary element method (BEM) is one of the numerical methods which is commonly used to solve the forward problem (FP) of electro-magnetic source imaging with realistic head geometries. Application of BEM generates large systems of linear equations with dense matrices. Generation and solution of these matrix equations are time and memory consuming. This study presents a relatively cheap and effective solution for parallel implementation of the BEM to reduce the processing times to clinically acceptable values. This is achieved using a parallel cluster of personal computers on a local area network. We used eight workstations and implemented a parallel version of the accelerated BEM approach that distributes the computation and the BEM matrix efficiently to the processors. The performance of the solver is evaluated in terms of the CPU operations and memory usage for different number of processors. Once the transfer matrix is computed, for a 12,294 node mesh, a single FP solution takes 676 ms on a single processor and 72 ms on eight processors. It was observed that workstation clusters are cost effective tools for solving the complex BEM models in a clinically acceptable time.

A Database Approach to Distributed State Space Generation

We study distributed state space generation on a cluster of workstations. It is explained why state space partitioning by a global hash function is problematic when states contain variables from unbounded domains, such as lists or other recursive datatypes. Our solution is to introduce a database which maintains a global numbering of state values. We also describe tree-compression, a technique of recursive state folding, and show that it is superior to manipulating plain state vectors.This solution is implemented and linked to the μCRL toolset, where state values are implemented as maximally shared terms (ATerms). However, it is applicable to other models as well, e.g., PROMELA models via the NIPS virtual machine. Our experiments show the trade-offs between keeping the database global, replicated, or local, depending on the available network bandwidth and latency.

Distributed Verification of Multi-threaded C++ Programs

Verification of multi-threaded C++ programs poses three major challenges: the large number of states, states with huge sizes, and time intensive expansions of states. This paper presents our efforts in addressing these issues by combining an efficient use of hard disk with the distribution of the state space on several computing nodes. The approach is applicable to clusters and multi-core machines with single or multiple hard disks. We exploit the concept of a signature of a state that allows the full state vector to stay on secondary memory. For a distributed exploration of the state space, we report the lessons learned from using different partitioning schemes, including Holzmann and Bosnacki's [G. Holzmann and D. Bosnacki. The design of a multi-core extension of the Spin Model Checker. IEEE Trans. on Software Engineering, 2007. To Appear] depth-slicing method, and their effects on blind and directed search algorithms.Empirical evaluation is done on our experimental C++ verification tool StEAM, which is capable of detecting errors such as segmentation faults, deadlocks, access conflicts, etc. The distributed algorithms are realized through MPI on a cluster of workstations.

A new modelling approach of wormhole-switched networks with finite buffers

Owing to its simple router design and insensitivity to the message distance, wormhole switching has been employed not only in networks for contemporary multicomputers but also, in networks for clusters of workstations, and more recently in networks for systems-on-chips. Although analytical performance models for wormhole-switched networks have been widely reported in the literature, most of these models have assumed no (or negligible) buffers at a given router in order to ease the derivation of the model. In this paper, we propose a new simple and yet accurate model to capture the effects of finite buffers on the performance of wormhole-switched networks. Simulation experiments demonstrate the validity of the suggested model under various operating conditions.

An Implementation of A Parallel Iterative Algorithm for the Solution of Large Banded System on A Cluster of Workstations

In this paper, we present a parallel iterative solution for large banded systems of linear equations based on incomplete LU-factorization (ILU). A master—workers parallel computing scheme is used. The proposed algorithm incurs reduced storage and communication overhead as compared to previous methods. The reduction in communication overhead has been achieved by pipelining the inter-workers message passing and limiting the communication between master and workers to initial work distribution and final results collection.The performance of the proposed algorithm is evaluated analytically and experimentally using Parallel Virtual Machine (PVM) as a parallel programming environment on a cluster of Linux work-stations. The obtained performance results show that the proposed algorithm exhibits higher efficiency for coarse grain computations corresponding to large matrix sizes and reduced number of processors.

Efficient mining of maximal frequent itemsets from databases on a cluster of workstations

In this paper, we propose a new algorithm, named Distributed Max-Miner (DMM), for mining maximal frequent itemsets from databases. A frequent itemset is maximal if none of its supersets is frequent. DMM requires very low communication and synchronization overhead in distributed computing systems. DMM has the local mining phase and the global mining phase. During the local mining phase, each node mines the local database to discover the local maximal frequent itemsets, then they form a set of maximal candidate itemsets for the top-down search in the subsequent global mining phase. A new prefix-tree data structure is developed to facilitate the storage and counting of the global candidate itemsets of different sizes. This global mining phase using the prefix-tree can work with any local mining algorithm. We implemented DMM on a cluster of workstations and evaluated its performance for various cases. DMM demonstrates better performance than other sequential and parallel algorithms, and its performance is quite scalable, even when there are large maximal frequent itemsets (i.e., long patterns) in databases.

Parallel algorithm for the solutions of PDEs in linux clustered workstations

In this paper we propose parallel algorithm for the solution of partial differential equations over a rectangular domain using the Crank–Nicholson method by cooperation with the DuFort–Frankel method and apply it on a model problem, namely, the heat conduction equation. One of the well known parallel techniques in solving partial differential equations in cluster computing environment is the domain decomposition technique. Using this technique, the whole domain is decomposed into subdomains, each of them has its own boundaries that are called the interface points. Parallelization is realized by approximating interface values using the unconditionally stable DuFort–Frankel explicit scheme, and these values serve as Neumann boundary conditions for the Crank–Nicholson implicit scheme in the subdomains. The numerical results show that our algorithm is more accurate than the algorithm based on the forward explicit method to approximate the values of the interface points, especially, when we use a small number of time steps. Moreover, these numerical results show that increasing the number of processors which are used in the cluster, yields an increase in the algorithm speedup.