Downloads & Free Reading Options - Results

We describe two main classes of one-sided trigonometric and hyperbolic Jacobi-type algorithms for computing eigenvalues and eigenvectors of Hermitian matrices. These types of algorithms exhibit significant advantages over many other eigenvalue algorithms. If the matrices permit, both types of algorithms compute the eigenvalues and eigenvectors with high relative accuracy. We present novel parallelization techniques for both trigonometric and hyperbolic classes of algorithms, as well as some new ideas on how pivoting in each cycle of the algorithm can improve the speed of the parallel one-sided algorithms. These parallelization approaches are applicable to both distributed-memory and shared-memory machines. The numerical testing performed indicates that the hyperbolic algorithms may be superior to the trigonometric ones, although, in theory, the latter seem more natural.

We design and implement a parallel algebraic multigrid method for isotropic graph Laplacian problems on multicore Graphical Processing Units (GPUs). The proposed AMG method is based on the aggregation framework. The setup phase of the algorithm uses a parallel maximal independent set algorithm in forming aggregates and the resulting coarse level hierarchy is then used in a K-cycle iteration solve phase with a $\ell^1$-Jacobi smoother. Numerical tests of a parallel implementation of the method for graphics processors are presented to demonstrate its effectiveness.

Multiple matching algorithms are used to locate the occurrences of patterns from a finite pattern set in a large input string. Aho-Corasick and Wu-Manber, two of the most well known algorithms for multiple matching require an increased computing power, particularly in cases where large-size datasets must be processed, as is common in computational biology applications. Over the past years, Graphics Processing Units (GPUs) have evolved to powerful parallel processors outperforming Central Processing Units (CPUs) in scientific calculations. Moreover, multiple GPUs can be used in parallel, forming hybrid computer cluster configurations to achieve an even higher processing throughput. This paper evaluates the speedup of the parallel implementation of the Aho-Corasick and Wu-Manber algorithms on a hybrid GPU cluster, when used to process a snapshot of the Expressed Sequence Tags of the human genome and for different problem parameters.

A new algorithm for the multiprocessor computation of the multidimensional discrete Fourier transform is presented. The algorithm eliminates the need for interprocessor communication and is highly scalable. The cost of eliminating interprocessor communication by this method is that one addition must be performed at each processing element for every input loaded into the machine. The algorithm is based on a new multidimensional discrete Fourier transform algorithm that computes a d-dimensional discrete Fourier transform by a set of independent k-dimensional discrete Fourier transforms (kd); it is a reduction algorithm in the sense that it has lowered the dimension of the Fourier transforms that are computed. The k-dimensional discrete Fourier transforms are performed on data derived from the input using only additions, and produce k-dimensional hyperplanes of the output array. Multidimensional discrete fourier transform algorithms, Parallel algorithms.

A new algorithm for the multiprocessor computation of the multidimensional discrete Fourier transform is presented. The algorithm eliminates the need for interprocessor communication and is highly scalable. The cost of eliminating interprocessor communication by this method is that one addition must be performed at each processing element for every input loaded into the machine. The algorithm is based on a new multidimensional discrete Fourier transform algorithm that computes a d-dimensional discrete Fourier transform by a set of independent k-dimensional discrete Fourier transforms (kd); it is a reduction algorithm in the sense that it has lowered the dimension of the Fourier transforms that are computed. The k-dimensional discrete Fourier transforms are performed on data derived from the input using only additions, and produce k-dimensional hyperplanes of the output array. Multidimensional discrete fourier transform algorithms, Parallel algorithms.

Several techniques to perform static and dynamic load balancing techniques for vision systems are presented. These techniques are novel in the sense that they capture the computational requirements of a task by examining the data when it is produced. Furthermore, they can be applied to many vision systems because many algorithms in different systems are either the same, or have similar computational characteristics. These techniques are evaluated by applying them on a parallel implementation of the algorithms in a motion estimation system on a hypercube multiprocessor system. The motion estimation system consists of the following steps: (1) extraction of features; (2) stereo match of images in one time instant; (3) time match of images from different time instants; (4) stereo match to compute final unambiguous points; and (5) computation of motion parameters. It is shown that the performance gains when these data decomposition and load balancing techniques are used are significant and the overhead of using these techniques is minimal.

Book Source: Digital Library of India Item 2015.193677 dc.contributor.author: S.veerabhadreswara Rao dc.date.accessioned: 2015-07-08T03:09:56Z dc.date.available: 2015-07-08T03:09:56Z dc.date.digitalpublicationdate: 2005-09-27 dc.identifier.barcode: 1990010091984 dc.identifier.origpath: /rawdataupload/upload/0091/984 dc.identifier.copyno: 1 dc.identifier.uri: http://www.new.dli.ernet.in/handle/2015/193677 dc.description.scannerno: 14 dc.description.scanningcentre: IIIT, Allahabad dc.description.main: 1 dc.description.tagged: 0 dc.description.totalpages: 48 dc.format.mimetype: application/pdf dc.language.iso: English dc.publisher: Indian Institute Of Technology Kanpur dc.rights: Out_of_copyright dc.source.library: Indian Institute Of Technology Kanpur dc.subject.classification: Technology dc.subject.classification: Engineering. Technology In General dc.subject.classification: Computer Science And Engineering dc.title: Efficient Parallel Algorithms For Some Weak External Visibility Problems

We study parallel comparison-based algorithms for finding all equivalence classes of a set of $n$ elements, where sorting according to some total order is not possible. Such scenarios arise, for example, in applications, such as in distributed computer security, where each of $n$ agents are working to identify the private group to which they belong, with the only operation available to them being a zero-knowledge pairwise-comparison (which is sometimes called a "secret handshake") that reveals only whether two agents are in the same group or in different groups. We provide new parallel algorithms for this problem, as well as new lower bounds and distribution-based analysis.

The purpose of this project has been to identify and test suitable technologies for implementation of fluid flow solvers -- possibly coupled with structures and heat equation solvers -- on MIMD parallel computers. In the course of this investigation much attention has been paid to efficient domain decomposition strategies for ADI-type algorithms. Multi-partitioning derives its efficiency from the assignment of several blocks of grid points to each processor in the parallel computer. A coarse-grain parallelism is obtained, and a near-perfect load balance results. In uni-partitioning every processor receives responsibility for exactly one block of grid points instead of several. This necessitates fine-grain pipelined program execution in order to obtain a reasonable load balance. Although fine-grain parallelism is less desirable on many systems, especially high-latency networks of workstations, uni-partition methods are still in wide use in production codes for flow problems. Consequently, it remains important to achieve good efficiency with this technique that has essentially been superseded by multi-partitioning for parallel ADI-type algorithms. Another reason for the concentration on improving the performance of pipeline methods is their applicability in other types of flow solver kernels with stronger implied data dependence. Analytical expressions can be derived for the size of the dynamic load imbalance incurred in traditional pipelines. From these it can be determined what is the optimal first-processor retardation that leads to the shortest total completion time for the pipeline process. Theoretical predictions of pipeline performance with and without optimization match experimental observations on the iPSC/860 very well. Analysis of pipeline performance also highlights the effect of uncareful grid partitioning in flow solvers that employ pipeline algorithms. If grid blocks at boundaries are not at least as large in the wall-normal direction as those immediately adjacent to them, then the first processor in the pipeline will receive a computational load that is less than that of subsequent processors, magnifying the pipeline slowdown effect. Extra compensation is needed for grid boundary effects, even if all grid blocks are equally sized.

The purpose of this project has been to identify and test suitable technologies for implementation of fluid flow solvers -- possibly coupled with structures and heat equation solvers -- on MIMD parallel computers. In the course of this investigation much attention has been paid to efficient domain decomposition strategies for ADI-type algorithms. Multi-partitioning derives its efficiency from the assignment of several blocks of grid points to each processor in the parallel computer. A coarse-grain parallelism is obtained, and a near-perfect load balance results. In uni-partitioning every processor receives responsibility for exactly one block of grid points instead of several. This necessitates fine-grain pipelined program execution in order to obtain a reasonable load balance. Although fine-grain parallelism is less desirable on many systems, especially high-latency networks of workstations, uni-partition methods are still in wide use in production codes for flow problems. Consequently, it remains important to achieve good efficiency with this technique that has essentially been superseded by multi-partitioning for parallel ADI-type algorithms. Another reason for the concentration on improving the performance of pipeline methods is their applicability in other types of flow solver kernels with stronger implied data dependence. Analytical expressions can be derived for the size of the dynamic load imbalance incurred in traditional pipelines. From these it can be determined what is the optimal first-processor retardation that leads to the shortest total completion time for the pipeline process. Theoretical predictions of pipeline performance with and without optimization match experimental observations on the iPSC/860 very well. Analysis of pipeline performance also highlights the effect of uncareful grid partitioning in flow solvers that employ pipeline algorithms. If grid blocks at boundaries are not at least as large in the wall-normal direction as those immediately adjacent to them, then the first processor in the pipeline will receive a computational load that is less than that of subsequent processors, magnifying the pipeline slowdown effect. Extra compensation is needed for grid boundary effects, even if all grid blocks are equally sized.

We obtain fast parallel algorithms for several scheduling problems. Some of the problems considered are: scheduling to minimize the number of tardy jobs; job sequencing with deadlines; scheduling to minimize earliness and tardiness penalties; channel assignment; and minimizing the mean finish time. The shared memory model of parallel computers is used. (Author)

We obtain fast parallel algorithms for several scheduling problems. Some of the problems considered are: scheduling to minimize the number of tardy jobs; job sequencing with deadlines; scheduling to minimize earliness and tardiness penalties; channel assignment; and minimizing the mean finish time. The shared memory model of parallel computers is used. (Author)

One of the major goals in the design of parallel processing machines and algorithms is to reduce the effects of the overhead introduced when a given problem is parallelized. A key contributor to overhead is communication time. Many architectures try to reduce this overhead by minimizing the actual time for communication, including latency and bandwidth. Another approach is to hide communication by overlapping it with computation. This paper presents the Proteus parallel computer and its effective use of communication hiding through overlapping communication/computation techniques. These techniques are easily extended for use in compiler support of parallel programming. We also address the complexity or rather simplicity, in achieving complete exchange on the Proteus Machine.

We present a high-level parallel calculus for nested sequences, NSC, offered as a possible theoretical "core" of an entire class of collection-oriented parallel languages. NSC is based on while-loops as opposed to general recursion. A formal, machine independent definition of the parallel time complexity and the work complexity of programs in NSC is given. Our main results are: (1) We give a translation method for a particular form of recursion, called map-recursion, into NSC, that preserves the time complexity and adds an arbitrarily small overhead to the work complexity, and (2) We give a compilation method for NSC into a very simple vector parallel machine, which preserves the time complexity and again adds an arbitrarily small overhead to the work complexity.

This paper considers the robust ${\cal D}$-stability margin problem under polynomic structured real parametric uncertainty. Based on the work of De Gaston and Safonov (1988), we have developed techniques such as, a parallel frequency sweeping strategy, different domain splitting schemes, which significantly reduce the computational complexity and guarantee the convergence.

This contract developed and disseminated papers, ideas, algorithms, analysis, software, applications, and implementations for parallel programming environments for computer vision and for vision applications. The work has been widely reported and highly influential. The most significant work centered on the Butterfly Parallel Processor, the MaxVideo pipelined parallel image processor, and the development of the real-time computer vision laboratory. For the Butterfly, the Psyche multi-model operating system was developed and the CONSUL autoparallelizing compiler was designed. Much basic and influential performance monitoring and debugging work was completed, resulting in working systems and novel algorithms. There was also significant research in systems and applications using other parallel architectures in the laboratory, such as the MaxVideo parallel pipelined image processor. The contract developed a heterogeneous parallel architecture involving pipelined and MIMD parallelism and integrated it with a robot head.

In this paper we solve on GPUs massive problems with large amount of data, which are not appropriate for solution with the SIMD technology. For the given problem we consider a three-level parallelization. The multithreading of CPU is used at the top level and graphic processors for massive computing. For solving problems of cluster analysis on GPUs the nearest neighbor method (NNM) is developed. This algorithm allows us to handle up to 2 millions records with number of features up to 25. Since sequential and parallel algorithms are fundamentally different, it is difficult to compare the computation times. However, some comparisons are made. The gain in the computing time is about 10 times. We plan to increase this factor up to 50-100 after fine tuning of algorithms.

We present DASH, a C++ template library that offers distributed data structures and parallel algorithms and implements a compiler-free PGAS (partitioned global address space) approach. DASH offers many productivity and performance features such as global-view data structures, efficient support for the owner-computes model, flexible multidimensional data distribution schemes and inter-operability with STL (standard template library) algorithms. DASH also features a flexible representation of the parallel target machine and allows the exploitation of several hierarchically organized levels of locality through a concept of Teams. We evaluate DASH on a number of benchmark applications and we port a scientific proxy application using the MPI two-sided model to DASH. We find that DASH offers excellent productivity and performance and demonstrate scalability up to 9800 cores.

Comparison of biological (DNA or protein) sequences provides insight into molecular structure, function, and homology, and is increasingly important as the available data bases become larger and more numerous. One method of increasing the speed of the calculations is to perform them in parallel. We present the results of initial investigations using the Intel iPSC hypercube and the Connection Machine for these comparisons. Since these machines have very different architectures, the issues and performance trade-offs discussed have a wide applicability for the parallel processing of biological sequence comparisons.

This thesis addresses a number of theoretical issues in parallel computation. There are many open questions relating to what can be done with parallel computers and what are the most effective techniques to use to develop parallel algorithms. The author examines various problems in hope of gaining insight to the general questions. One topic that is investigated is the relationship between sequential and parallel algorithms. Introduced is the concept of a P-complete algorithm to capture what it means for an algorithm to be inherently sequential . It is shown that a number of sequential greedy algorithms are P-complete, including the greedy algorithm for finding a path in a graph. However, a problem is not necessarily difficult if an algorithm to solve it is P-complete. In some cases, the natural sequential algorithm is P-complete but a different technique gives a fast parallel algorithm. This shows that it is necessary to use different techniques for parallel computation than are used for sequential computation. Fast parallel algorithms for a number of simple graph theory problems are given. The algorithms illustrate a number of different techniques that are useful for parallel algorithms. The final topic that we address is parallel approximation of P-complete problems.

We analyze the effects of hardware faults on the performance of computer-implemented signal detectors, as measured by the probability of detection and the probability of false alarm. We then use these results to design fault-tolerant detectors using hardware redundancy. Keywords include. Fault tolerance, Signal Processing, and Parallel Architecture.

In this Project, the main objective was to develop multiresolution wavelet algorithm to study the flame acceleration and to understand the mechanism of the transition of deflagration to detonation. In the first half year of the project. The author completed the study of two dimensional detonation waves based on hybrid high order methods (a combination of high order Essentially Non-oscillatory (ENO) methods and spectral methods and Shock Tracking methods). The result on the detonation waves was published in the AIAA Journal. In the remaining time of the two years, first, the author completed the theoretical and algorithmic studies of the adaptive wavelet method, which could handle nonperodic boundary conditions and nonlinear time dependent PDE's. Secondly, the author implemented the adaptive wavelet methods for the solution of one-dimensional flame propagation; thirdly, the author developed a Fortran code WL2D (more than 13,000 lines) for the two dimensional multi-scale wavelet algorithms with an efficient data structure and implemented a second order implicit factorized scheme for the adaptive wavelet methods.

In this Project, the main objective was to develop multiresolution wavelet algorithm to study the flame acceleration and to understand the mechanism of the transition of deflagration to detonation. In the first half year of the project. The author completed the study of two dimensional detonation waves based on hybrid high order methods (a combination of high order Essentially Non-oscillatory (ENO) methods and spectral methods and Shock Tracking methods). The result on the detonation waves was published in the AIAA Journal. In the remaining time of the two years, first, the author completed the theoretical and algorithmic studies of the adaptive wavelet method, which could handle nonperodic boundary conditions and nonlinear time dependent PDE's. Secondly, the author implemented the adaptive wavelet methods for the solution of one-dimensional flame propagation; thirdly, the author developed a Fortran code WL2D (more than 13,000 lines) for the two dimensional multi-scale wavelet algorithms with an efficient data structure and implemented a second order implicit factorized scheme for the adaptive wavelet methods.

As a Guest Computational Investigator under the NASA administered component of the High Performance Computing and Communication Program, we implemented a massively parallel genetic algorithm on the MasPar SIMD computer. Experiments were conducted using Earth Science data in the domains of meteorology and oceanography. Results obtained in these domains are competitive with, and in most cases better than, similar problems solved using other methods. In the meteorological domain, we chose to identify clouds using AVHRR spectral data. Four cloud speciations were used although most researchers settle for three. Results were remarkedly consistent across all tests (91 accuracy). Refinements of this method may lead to more timely and complete information for Global Circulation Models (GCMS) that are prevalent in weather forecasting and global environment studies. In the oceanographic domain, we chose to identify ocean currents from a spectrometer having similar characteristics to AVHRR. Here the results were mixed (60 to 80 accuracy). Given that one is willing to run the experiment several times (say 10), then it is acceptable to claim the higher accuracy rating. This problem has never been successfully automated. Therefore, these results are encouraging even though less impressive than the cloud experiment. Successful conclusion of an automated ocean current detection system would impact coastal fishing, naval tactics, and the study of micro-climates. Finally we contributed to the basic knowledge of GA (genetic algorithm) behavior in parallel environments. We developed better knowledge of the use of subpopulations in the context of shared breeding pools and the migration of individuals. Rigorous experiments were conducted based on quantifiable performance criteria. While much of the work confirmed current wisdom, for the first time we were able to submit conclusive evidence. The software developed under this grant was placed in the public domain. An extensive user's manual was written and distributed nationwide to scientists whose work might benefit from its availability. Several papers, including two journal articles, were produced.

As a Guest Computational Investigator under the NASA administered component of the High Performance Computing and Communication Program, we implemented a massively parallel genetic algorithm on the MasPar SIMD computer. Experiments were conducted using Earth Science data in the domains of meteorology and oceanography. Results obtained in these domains are competitive with, and in most cases better than, similar problems solved using other methods. In the meteorological domain, we chose to identify clouds using AVHRR spectral data. Four cloud speciations were used although most researchers settle for three. Results were remarkedly consistent across all tests (91 accuracy). Refinements of this method may lead to more timely and complete information for Global Circulation Models (GCMS) that are prevalent in weather forecasting and global environment studies. In the oceanographic domain, we chose to identify ocean currents from a spectrometer having similar characteristics to AVHRR. Here the results were mixed (60 to 80 accuracy). Given that one is willing to run the experiment several times (say 10), then it is acceptable to claim the higher accuracy rating. This problem has never been successfully automated. Therefore, these results are encouraging even though less impressive than the cloud experiment. Successful conclusion of an automated ocean current detection system would impact coastal fishing, naval tactics, and the study of micro-climates. Finally we contributed to the basic knowledge of GA (genetic algorithm) behavior in parallel environments. We developed better knowledge of the use of subpopulations in the context of shared breeding pools and the migration of individuals. Rigorous experiments were conducted based on quantifiable performance criteria. While much of the work confirmed current wisdom, for the first time we were able to submit conclusive evidence. The software developed under this grant was placed in the public domain. An extensive user's manual was written and distributed nationwide to scientists whose work might benefit from its availability. Several papers, including two journal articles, were produced.

As a Guest Computational Investigator under the NASA administered component of the High Performance Computing and Communication Program, we implemented a massively parallel genetic algorithm on the MasPar SIMD computer. Experiments were conducted using Earth Science data in the domains of meteorology and oceanography. Results obtained in these domains are competitive with, and in most cases better than, similar problems solved using other methods. In the meteorological domain, we chose to identify clouds using AVHRR spectral data. Four cloud speciations were used although most researchers settle for three. Results were remarkedly consistent across all tests (91 accuracy). Refinements of this method may lead to more timely and complete information for Global Circulation Models (GCMS) that are prevalent in weather forecasting and global environment studies. In the oceanographic domain, we chose to identify ocean currents from a spectrometer having similar characteristics to AVHRR. Here the results were mixed (60 to 80 accuracy). Given that one is willing to run the experiment several times (say 10), then it is acceptable to claim the higher accuracy rating. This problem has never been successfully automated. Therefore, these results are encouraging even though less impressive than the cloud experiment. Successful conclusion of an automated ocean current detection system would impact coastal fishing, naval tactics, and the study of micro-climates. Finally we contributed to the basic knowledge of GA (genetic algorithm) behavior in parallel environments. We developed better knowledge of the use of subpopulations in the context of shared breeding pools and the migration of individuals. Rigorous experiments were conducted based on quantifiable performance criteria. While much of the work confirmed current wisdom, for the first time we were able to submit conclusive evidence. The software developed under this grant was placed in the public domain. An extensive user's manual was written and distributed nationwide to scientists whose work might benefit from its availability. Several papers, including two journal articles, were produced.

The Bethe-Salpeter eigenvalue problem is a dense structured eigenvalue problem arising from discretized Bethe-Salpeter equation in the context of computing exciton energies and states. A computational challenge is that at least half of the eigenvalues and the associated eigenvectors are desired in practice. We establish the equivalence between Bethe-Salpeter eigenvalue problems and real Hamiltonian eigenvalue problems. Based on theoretical analysis, structure preserving algorithms for a class of Bethe-Salpeter eigenvalue problems are proposed. We also show that for this class of problems all eigenvalues obtained from the Tamm-Dancoff approximation are overestimated. In order to solve large scale problems of practical interest, we discuss parallel implementations of our algorithms targeting distributed memory systems. Several numerical examples are presented to demonstrate the efficiency and accuracy of our algorithms.

The training of Support Vector Machines may be a very difficult task when dealing with very large datasets. The memory requirement and the time consumption of the SVMs algorithms grow rapidly with the increase of the data. To overcome these drawbacks, we propose a parallel decomposition algorithmic scheme for SVMs training for which we prove global convergence under suitable conditions. We outline how these assumptions can be satisfied in practice and we suggest various specific implementations exploiting the adaptable structure of the algorithmic model.

Parallel least mean square-partial parallel interference cancelation (PLMS-PPIC) is a partial interference cancelation which employs adaptive multistage structure. In this algorithm the channel phases for all users are assumed to be known. Having only their quarters in (0,2\pi), a modified version of PLMS-PPIC is proposed in this paper to simultaneously estimate the channel phases and the cancelation weights. Simulation examples are given in the cases of balanced, unbalanced and time varying channels to show the performance of the modified PLMS-PPIC method.

The scalability of massively parallel algorithms is a fundamental question in computer science. We study the scalability and the efficiency of a conservative massively parallel algorithm for discrete-event simulations where the discrete events are Poisson arrivals. The parallel algorithm is applicable to a wide range of problems, including dynamic Monte Carlo simulations for large asynchronous systems with short-range interactions. The evolution of the simulated time horizon is analogous to a growing and fluctuating surface, and the efficiency of the algorithm corresponds to the density of local minima of this surface. In one dimension we find that the steady state of the macroscopic landscape is governed by the Edwards-Wilkinson Hamiltonian, which implies that the algorithm is scalable. Preliminary results for higher-dimensional logical topologies are discussed.

The computational complexity of internal diffusion-limited aggregation (DLA) is examined from both a theoretical and a practical point of view. We show that for two or more dimensions, the problem of predicting the cluster from a given set of paths is complete for the complexity class CC, the subset of P characterized by circuits composed of comparator gates. CC-completeness is believed to imply that, in the worst case, growing a cluster of size n requires polynomial time in n even on a parallel computer. A parallel relaxation algorithm is presented that uses the fact that clusters are nearly spherical to guess the cluster from a given set of paths, and then corrects defects in the guessed cluster through a non-local annihilation process. The parallel running time of the relaxation algorithm for two-dimensional internal DLA is studied by simulating it on a serial computer. The numerical results are compatible with a running time that is either polylogarithmic in n or a small power of n. Thus the computational resources needed to grow large clusters are significantly less on average than the worst-case analysis would suggest. For a parallel machine with k processors, we show that random clusters in d dimensions can be generated in O((n/k + log k) n^{2/d}) steps. This is a significant speedup over explicit sequential simulation, which takes O(n^{1+2/d}) time on average. Finally, we show that in one dimension internal DLA can be predicted in O(log n) parallel time, and so is in the complexity class NC.

Computer vision is regarded as one of the most complex and computationally intensive problems. An integrated vision system (IVS) is a system that uses vision algorithms from all levels of processing to perform for a high level application (e.g., object recognition). An IVS normally involves algorithms from low level, intermediate level, and high level vision. Designing parallel architectures for vision systems is of tremendous interest to researchers. Several issues are addressed in parallel architectures and parallel algorithms for integrated vision systems.

In view of the tremendous computing power jump of modern RISC processors the interest in parallel computing seems to be thinning out. Why use a complicated system of parallel processors, if the problem can be solved by a single powerful micro-chip? It is a general law, however, that exponential growth will always end by some kind of a saturation, and then parallelism will again become a hot topic. We try to prepare ourselves for this eventuality. The MPPC project started in 1990 in the heydeys of parallelism and produced four ASTRA machines (presented at CHEP'92) with 4k processors (which are expandable to 16k) based on `yesterday's chip-technology' (chip presented at CHEP'91). These machines now provide excellent test-beds for algorithmic developments in a complete, real environment. We are developing for example fast-pattern recognition algorithms which could be used in high-energy physics experiments at the LHC (planned to be operational after 2004 at CERN) for triggering and data reduction. The basic feature of our ASP (Associative String Processor) approach is to use extremely simple (thus very cheap) processor elements but in huge quantities (up to millions of processors) connected together by a very simple string-like communication chain. In this paper we present powerful algorithms based on this architecture indicating the performance perspectives if the hardware (i.e. chip fabrication) quality reaches `present or even future technology levels'.

Tensor factorization has proven useful in a wide range of applications, from sensor array processing to communications, speech and audio signal processing, and machine learning. With few recent exceptions, all tensor factorization algorithms were originally developed for centralized, in-memory computation on a single machine; and the few that break away from this mold do not easily incorporate practically important constraints, such as nonnegativity. A new constrained tensor factorization framework is proposed in this paper, building upon the Alternating Direction method of Multipliers (ADMoM). It is shown that this simplifies computations, bypassing the need to solve constrained optimization problems in each iteration; and it naturally leads to distributed algorithms suitable for parallel implementation on regular high-performance computing (e.g., mesh) architectures. This opens the door for many emerging big data-enabled applications. The methodology is exemplified using nonnegativity as a baseline constraint, but the proposed framework can more-or-less readily incorporate many other types of constraints. Numerical experiments are very encouraging, indicating that the ADMoM-based nonnegative tensor factorization (NTF) has high potential as an alternative to state-of-the-art approaches.

Phase I is complete for the development of a Computational Fluid Dynamics parallel code with automatic grid generation and adaptation for the Euler analysis of flow over complex geometries. SPLITFLOW, an unstructured Cartesian grid code developed at Lockheed Martin Tactical Aircraft Systems, has been modified for a distributed memory/massively parallel computing environment. The parallel code is operational on an SGI network, Cray J90 and C90 vector machines, SGI Power Challenge, and Cray T3D and IBM SP2 massively parallel machines. Parallel Virtual Machine (PVM) is the message passing protocol for portability to various architectures. A domain decomposition technique was developed which enforces dynamic load balancing to improve solution speed and memory requirements. A host/node algorithm distributes the tasks. The solver parallelizes very well, and scales with the number of processors. Partially parallelized and non-parallelized tasks consume most of the wall clock time in a very fine grain environment. Timing comparisons on a Cray C90 demonstrate that Parallel SPLITFLOW runs 2.4 times faster on 8 processors than its non-parallel counterpart autotasked over 8 processors.

We provide a convergence analysis for a new fractional time-stepping technique for the incompressible Navier-Stokes equations based on direction splitting. This new technique is of linear complexity, unconditionally stable and convergent, and suitable for massive parallelization.

This paper presents a pragmatic comparison of three parallel algorithms for finding connected components. together with optimizations on these algorithms. Those being compared are two similar algorithms by Awerbuch and Shiloach 2 and by Shiloach and Vishkin 19 and a randomized contraction algorithm by Blelloch 7, based on algorithms by Reif 18 and Phillips 17. Major improvements are given for the first two which significantly reduces the super- linear component of their work complexity. An improvement is also given for randomized algorithm. and this algorithm, and this to be the fastest of those tested. These comparisons are presented with NESL data-parallel code is executed on a Connection Machine 2.

These are the lecture notes for CS 15-840B, a hands-on class in programming parallel algorithms. The class was taught in the fall of 1992 by Guy Blelloch, using the programming, language NESL. It stressed the clean and concise expression of a variety of parallel algorithms. About 35 graduate students attended the class', of whom 28 took it for credit. These notes were written by students in the class, and were then reviewed and organized by Guy Blelloch and Jonathan Hardwick. The sample NESL code has been converted from the older LISP-style syntax into the new NlL-style syntax. These notes are not in a polished form, and probably contain several errors and omissions, particularly with respect to references in the literature. Corrections are welcome.

Given a reduced affine algebra A over a perfect field K, we present parallel algorithms to compute the normalization \bar{A} of A. Our starting point is the algorithm of Greuel, Laplagne, and Seelisch, which is an improvement of de Jong's algorithm. First, we propose to stratify the singular locus Sing(A) in a way which is compatible with normalization, apply a local version of the normalization algorithm at each stratum, and find \bar{A} by putting the local results together. Second, in the case where K = Q is the field of rationals, we propose modular versions of the global and local-to-global algorithms. We have implemented our algorithms in the computer algebra system SINGULAR and compare their performance with that of the algorithm of Greuel, Laplagne, and Seelisch. In the case where K = Q, we also discuss the use of modular computations of Groebner bases, radicals, and primary decompositions. We point out that in most examples, the new algorithms outperform the algorithm of Greuel, Laplagne, and Seelisch by far, even if we do not run them in parallel.

Given a reduced affine algebra A over a perfect field K, we present parallel algorithms to compute the normalization \bar{A} of A. Our starting point is the algorithm of Greuel, Laplagne, and Seelisch, which is an improvement of de Jong's algorithm. First, we propose to stratify the singular locus Sing(A) in a way which is compatible with normalization, apply a local version of the normalization algorithm at each stratum, and find \bar{A} by putting the local results together. Second, in the case where K = Q is the field of rationals, we propose modular versions of the global and local-to-global algorithms. We have implemented our algorithms in the computer algebra system SINGULAR and compare their performance with that of the algorithm of Greuel, Laplagne, and Seelisch. In the case where K = Q, we also discuss the use of modular computations of Groebner bases, radicals, and primary decompositions. We point out that in most examples, the new algorithms outperform the algorithm of Greuel, Laplagne, and Seelisch by far, even if we do not run them in parallel.

Given a reduced affine algebra A over a perfect field K, we present parallel algorithms to compute the normalization \bar{A} of A. Our starting point is the algorithm of Greuel, Laplagne, and Seelisch, which is an improvement of de Jong's algorithm. First, we propose to stratify the singular locus Sing(A) in a way which is compatible with normalization, apply a local version of the normalization algorithm at each stratum, and find \bar{A} by putting the local results together. Second, in the case where K = Q is the field of rationals, we propose modular versions of the global and local-to-global algorithms. We have implemented our algorithms in the computer algebra system SINGULAR and compare their performance with that of the algorithm of Greuel, Laplagne, and Seelisch. In the case where K = Q, we also discuss the use of modular computations of Groebner bases, radicals, and primary decompositions. We point out that in most examples, the new algorithms outperform the algorithm of Greuel, Laplagne, and Seelisch by far, even if we do not run them in parallel.

Given a reduced affine algebra A over a perfect field K, we present parallel algorithms to compute the normalization \bar{A} of A. Our starting point is the algorithm of Greuel, Laplagne, and Seelisch, which is an improvement of de Jong's algorithm. First, we propose to stratify the singular locus Sing(A) in a way which is compatible with normalization, apply a local version of the normalization algorithm at each stratum, and find \bar{A} by putting the local results together. Second, in the case where K = Q is the field of rationals, we propose modular versions of the global and local-to-global algorithms. We have implemented our algorithms in the computer algebra system SINGULAR and compare their performance with that of the algorithm of Greuel, Laplagne, and Seelisch. In the case where K = Q, we also discuss the use of modular computations of Groebner bases, radicals, and primary decompositions. We point out that in most examples, the new algorithms outperform the algorithm of Greuel, Laplagne, and Seelisch by far, even if we do not run them in parallel.

Given a reduced affine algebra A over a perfect field K, we present parallel algorithms to compute the normalization \bar{A} of A. Our starting point is the algorithm of Greuel, Laplagne, and Seelisch, which is an improvement of de Jong's algorithm. First, we propose to stratify the singular locus Sing(A) in a way which is compatible with normalization, apply a local version of the normalization algorithm at each stratum, and find \bar{A} by putting the local results together. Second, in the case where K = Q is the field of rationals, we propose modular versions of the global and local-to-global algorithms. We have implemented our algorithms in the computer algebra system SINGULAR and compare their performance with that of the algorithm of Greuel, Laplagne, and Seelisch. In the case where K = Q, we also discuss the use of modular computations of Groebner bases, radicals, and primary decompositions. We point out that in most examples, the new algorithms outperform the algorithm of Greuel, Laplagne, and Seelisch by far, even if we do not run them in parallel.

Significant advances have been made recently toward understanding the properties of materials through theoretical approaches. These approaches are based either on first-principles quantum mechanical formulations or semi-empirical formulations, and have benefitted from increases in computational power. The advent of parallel computing has propelled the theoretical approaches to a new level of realism in modeling physical systems of interest. The theoretical methods and simulation techniques that are currently under development are bound to become powerful tools in understanding, exploring and predicting the properties of existing and novel materials. The aim of this symposium was to bring together scientists from several subfields of materials theory and simulations. Its purpose was: to make contact with traditional continuum approaches to materials theory; to discuss critically current developments in computational and simulational approaches specifically aimed at addressing real materials problems, and with emphasis on parallel computing; and to present examples of the most successful applications of computational and simulational work to date.

We describe a hardware upgrade that was performed on the nCUBE 2 massively parallel computer at MIT's Earth Resources Laboratory. The upgrade involved the installation of additional memory, additional disk space, and an additional small nCUBE 2 computer. The reason for performing the upgrade was to provide ERL with a computational environment suitable for conducting large-scale 2-D and 3-D wave propagation simulations. We also describe two recent projects that were performed on the upgraded nCUBE 2. One project was a 3-D simulation of earthquake-generated ground motion in the Boston Basin. The other project was the development of a parallel irregular-grid modeling code which is well-suited for 2-D modeling of regional wave propagation in the presence of surface topography.

A tournament is a digraph T=( V, E) in which, for every pair of vertices, u & v, exactly one of (u, v), (v, u) is in E. Two classical theorems about tournaments are that every tournament has a Hamiltonian path, and every strongly connected tournament has a Hamiltonian cycle. Furthermore, it is known how to find these in polynomial time. In this paper we discuss the parallel complexity of these problems. Our main result is that constructing a Hamiltonian path in a general tournament and a Hamiltonian cycle in a strongly connected tournament are both in NC. In addition, we give an NC algorithm for finding a Hamiltonian path with one fixed endpoint. In finding fast parallel algorithms, we also obtain new proofs for the theorems.

We obtain fast parallel algorithms for several scheduling problems. Some of the problems considered are: scheduling to minimize the number of tardy jobs; job sequencing with deadlines; scheduling to minimize earliness and tardiness penalties; channel assignment; and minimizing the mean finish time. The shared memory model of parallel computers is used. (Author)

The determination of the fundamental (lowest) natural vibration frequencies and associated mode shapes is a key step used to uncover and correct potential failures or problem areas in most complex structures. However, the computation time taken by finite element codes to evaluate these natural frequencies is significant, often the most computationally intensive part of structural analysis calculations. There is continuing need to reduce this computation time. This study addresses this need by developing methods for parallel computation.

The main goal of this project was efficient distributed parallel and workstation cluster implementations of Newton-Krylov-Schwarz (NKS) solvers for implicit Computational Fluid Dynamics (CFD.) "Newton" refers to a quadratically convergent nonlinear iteration using gradient information based on the true residual, "Krylov" to an inner linear iteration that accesses the Jacobian matrix only through highly parallelizable sparse matrix-vector products, and "Schwarz" to a domain decomposition form of preconditioning the inner Krylov iterations with primarily neighbor-only exchange of data between the processors. Prior experience has established that Newton-Krylov methods are competitive solvers in the CFD context and that Krylov-Schwarz methods port well to distributed memory computers. The combination of the techniques into Newton-Krylov-Schwarz was implemented on 2D and 3D unstructured Euler codes on the parallel testbeds that used to be at LaRC and on several other parallel computers operated by other agencies or made available by the vendors. Early implementations were made directly in Massively Parallel Integration (MPI) with parallel solvers we adapted from legacy NASA codes and enhanced for full NKS functionality. Later implementations were made in the framework of the PETSC library from Argonne National Laboratory, which now includes pseudo-transient continuation Newton-Krylov-Schwarz solver capability (as a result of demands we made upon PETSC during our early porting experiences). A secondary project pursued with funding from this contract was parallel implicit solvers in acoustics, specifically in the Helmholtz formulation. A 2D acoustic inverse problem has been solved in parallel within the PETSC framework.

We give a new randomized parallel RAM algorithm for finding a spanning forest of an undirected graph in logarithmic time. These time bounds hold with arbitrary high probability for any input graph (i.e., we do not assume random input; these bounds hold for the worst case input graph). This result assumes a parallel RAM model which allows both concurrent writes and concurrent reads. Furthermore, we show that if the graph is not very sparse (i.e., if the number of edges is at least a logarithmic squared factor more than the number of vertices) than we can achieve a linear processor time product (even for logarithmic time bounds) for finding a spanning tree--which is optimal for the parallel RAM model. Furthermore, we can also achieve a linear processor, time product for even sparser graphs with only slight time increase. Keywords include: graph connectivity, parallel algorithms, optimal algorithms, randomized algorithms.