Downloads & Free Reading Options - Results

This research concerns the algorithmic development, computer implementation and direct numerical simulation of incompressible and compressible flow of naval relevance. Calculations were executed on a class of current generation multiprocessors. Pseudospectral methods were used exclusively. Lack of parallel algorithms critical to the effective implementation of spectral methods on parallel computers necessitated the need for the development of parallel FFT algorithms for real, conjugate symmetric and real symmetric sequences. These algorithms are applied to spectral methods, but also in many areas of scientific computing. The last algorithm, the parallel fast discrete cosine transform, is used extensively in image and signal processing. The parallel Fourier pseudospectral method for the incompressible Navier-Stokes equations was developed and implemented on many multiprocessors. Reconnection of orthogonally interacting vortex tubes was then investigated using the algorithm on parallel computers as well as vector supercomputers. The parallel Fourier pseudospectral method for the compressible Navier-Stokes equations was also developed. Shock / vortex interactions in two dimensions were investigated.

This research concerns the algorithmic development, computer implementation and direct numerical simulation of incompressible and compressible flow of naval relevance. Calculations were executed on a class of current generation multiprocessors. Pseudospectral methods were used exclusively. Lack of parallel algorithms critical to the effective implementation of spectral methods on parallel computers necessitated the need for the development of parallel FFT algorithms for real, conjugate symmetric and real symmetric sequences. These algorithms are applied to spectral methods, but also in many areas of scientific computing. The last algorithm, the parallel fast discrete cosine transform, is used extensively in image and signal processing. The parallel Fourier pseudospectral method for the incompressible Navier-Stokes equations was developed and implemented on many multiprocessors. Reconnection of orthogonally interacting vortex tubes was then investigated using the algorithm on parallel computers as well as vector supercomputers. The parallel Fourier pseudospectral method for the compressible Navier-Stokes equations was also developed. Shock / vortex interactions in two dimensions were investigated.

The ability to express a program as a hierarchical composition of parts is an essential tool in managing the complexity of software and a key abstraction this provides is to separate the representation of data from the computation. Many current parallel programming models use a shared memory model to provide data abstraction but this doesn't scale well with large numbers of cores due to non-determinism and access latency. This paper proposes a simple programming model that allows scalable parallel programs to be expressed with distributed representations of data and it provides the programmer with the flexibility to employ shared or distributed styles of data-parallelism where applicable. It is capable of an efficient implementation, and with the provision of a small set of primitive capabilities in the hardware, it can be compiled to operate directly on the hardware, in the same way stack-based allocation operates for subroutines in sequential machines.

Multi-core architectures feature an intricate hierarchy of cache memories, with multiple levels and sizes. To adequately decompose an application according to the traits of a particular memory hierarchy is a cumbersome task that may be rewarded with significant performance gains. The current state-of-the-art in memory-hierarchy-aware parallel computing delegates this endeavour on the programmer, demanding from him deep knowledge of both parallel programming and computer architecture. In this paper we propose the shifting of these memory-hierarchy-related concerns to the run-time system, which then takes on the responsibility of distributing the computation's data across the target memory hierarchy. We evaluate our approach from a performance perspective, comparing it against the common cache-neglectful data decomposition strategy.

v. p. 28 cm

We present a set of programming tools (classes and functions written in C++ and based on Message Passing Interface) for fast development of generic parallel (and non-parallel) lattice simulations. They are collectively called MDP 1.2. These programming tools include classes and algorithms for matrices, random number generators, distributed lattices (with arbitrary topology), fields and parallel iterations. No previous knowledge of MPI is required in order to use them. Some applications in electromagnetism, electronics, condensed matter and lattice QCD are presented.

We present a set of programming tools (classes and functions written in C++ and based on Message Passing Interface) for fast development of generic parallel (and non-parallel) lattice simulations. They are collectively called MDP 1.2. These programming tools include classes and algorithms for matrices, random number generators, distributed lattices (with arbitrary topology), fields and parallel iterations. No previous knowledge of MPI is required in order to use them. Some applications in electromagnetism, electronics, condensed matter and lattice QCD are presented.

Matrix computations, especially iterative PDE solving (and the sparse matrix vector multiplication subproblem within) using conjugate gradient algorithm, and LU/Cholesky decomposition for solving system of linear equations, form the kernel of many applications, such as circuit simulators, computational fluid dynamics or structural analysis etc. The problem of designing approaches for parallelizing these computations, to get good speedups as much as possible as per Amdahl's law, has been continuously researched upon. In this paper, we discuss approaches based on the use of finite projective geometry graphs for these two problems. For the problem of conjugate gradient algorithm, the approach looks at an alternative data distribution based on projective-geometry concepts. It is proved that this data distribution is an optimal data distribution for scheduling the main problem of dense matrix-vector multiplication. For the problem of parallel LU/Cholesky decomposition of general matrices, the approach is motivated by the recently published scheme for interconnects of distributed systems, perfect difference networks. We find that projective-geometry based graphs indeed offer an exciting way of parallelizing these computations, and in fact many others. Moreover, their applications ranges from architectural ones (interconnect choice) to algorithmic ones (data distributions).

Explicit codes are often used to simulate the nonlinear dynamics of large-scale structural systems, even for low frequency response, because the storage and CPU requirements entailed by the repeated factorizations traditionally found in implicit codes rapidly overwhelm the available computing resources. With the advent of parallel processing, this trend is accelerating because explicit schemes are also easier to parallellize than implicit ones. However, the time step restriction imposed by the Courant stability condition on all explicit schemes cannot yet and perhaps will never be offset by the speed of parallel hardware. Therefore, it is essential to develop efficient and robust alternatives to direct methods that are also amenable to massively parallel processing because implicit codes using unconditionally stable time-integration algorithms are computationally more efficient than explicit codes when simulating low-frequency dynamics. Here we present a domain decomposition method for implicit schemes that requires significantly less storage than factorization algorithms, that is several times faster than other popular direct and iterative methods, that can be easily implemented on both shared and local memory parallel processors, and that is both computationally and communication-wise efficient. The proposed transient domain decomposition method is an extension of the method of Finite Element Tearing and Interconnecting (FETI) developed by Farhat and Roux for the solution of static problems. Serial and parallel performance results on the CRAY Y-MP/8 and the iPSC-860/128 systems are reported and analyzed for realistic structural dynamics problems. These results establish the superiority of the FETI method over both the serial/parallel conjugate gradient algorithm with diagonal scaling and the serial/parallel direct method, and contrast the computational power of the iPSC-860/128 parallel processor with that of the CRAY Y-MP/8 system.

Explicit codes are often used to simulate the nonlinear dynamics of large-scale structural systems, even for low frequency response, because the storage and CPU requirements entailed by the repeated factorizations traditionally found in implicit codes rapidly overwhelm the available computing resources. With the advent of parallel processing, this trend is accelerating because explicit schemes are also easier to parallellize than implicit ones. However, the time step restriction imposed by the Courant stability condition on all explicit schemes cannot yet and perhaps will never be offset by the speed of parallel hardware. Therefore, it is essential to develop efficient and robust alternatives to direct methods that are also amenable to massively parallel processing because implicit codes using unconditionally stable time-integration algorithms are computationally more efficient than explicit codes when simulating low-frequency dynamics. Here we present a domain decomposition method for implicit schemes that requires significantly less storage than factorization algorithms, that is several times faster than other popular direct and iterative methods, that can be easily implemented on both shared and local memory parallel processors, and that is both computationally and communication-wise efficient. The proposed transient domain decomposition method is an extension of the method of Finite Element Tearing and Interconnecting (FETI) developed by Farhat and Roux for the solution of static problems. Serial and parallel performance results on the CRAY Y-MP/8 and the iPSC-860/128 systems are reported and analyzed for realistic structural dynamics problems. These results establish the superiority of the FETI method over both the serial/parallel conjugate gradient algorithm with diagonal scaling and the serial/parallel direct method, and contrast the computational power of the iPSC-860/128 parallel processor with that of the CRAY Y-MP/8 system.

Parallel overset grid assembly techniques have recently been implemented in the U.S. Army Helios software for rotorcraft aeromechanics simulations. The work presented in this report describes an innovative active load balancing algorithm that improves the robustness and scalability of domain connectivity operations. Another aspect of this works in to development of conservative overset grid methods. Preliminary results in 2-D for cell-centered unstructured grid show feasibility of this approach and improvements in predictions when compared with conventional overset grid approaches.

Comment in response to Phys. Rev. Lett. 112, 046401 (2014)

Particle-in-cell methods couple mesh-based methods for the solution of continuum mechanics problems, with the ability to advect and evolve particles. They have a long history and many applications in scientific computing. However, they have most often only been implemented for either sequential codes, or parallel codes with static meshes that are statically partitioned. In contrast, many mesh-based codes today use adaptively changing, dynamically partitioned meshes, and can scale to thousands or tens of thousands of processors. Consequently, there is a need to revisit the data structures and algorithms necessary to use particle methods with modern, mesh-based methods. Here we review commonly encountered requirements of particle-in-cell methods, and describe efficient ways to implement them in the context of large-scale parallel finite-element codes that use dynamically changing meshes. We also provide practical experience for how to address bottlenecks that impede the efficient implementation of these algorithms and demonstrate with numerical tests both that our algorithms can be implemented with optimal complexity and that they are suitable for very large-scale, practical applications. We provide a reference implementation in ASPECT, an open source code for geodynamic mantle-convection simulations built on the deal.II library.

The equations of motion for structures with adaptive elements for vibration control are presented for parallel computations to be used as a software package for real-time control of flexible space structures. A brief introduction of the state-of-the-art parallel computational capability is also presented. Time marching strategies are developed for an effective use of massive parallel mapping, partitioning, and the necessary arithmetic operations. An example is offered for the simulation of control-structure interaction on a parallel computer and the impact of the approach presented for applications in other disciplines than aerospace industry is assessed.

Many computational problems in image processing, signal processing, and scientific computing are naturally structured for either pipelined or parallel computation. When mapping such problems onto a parallel architecture it is often necessary to aggregate an obvious problem decomposition. Even in this context the general mapping problem is known to be computationally intractable, but recent advances have been made in identifying classes of problems and architectures for which optimal solutions can be found in polynomial time. Among these, the mapping of pipelined or parallel computations onto linear array, shared memory, and host-satellite systems figures prominently. This paper extends that work first by showing how to improve existing serial mapping algorithms. These improvements have significantly lower time and space complexities: in one case a published O(nm sup 3) time algorithm for mapping m modules onto n processors is reduced to an O(nm log m) time complexity, and its space requirements reduced from O(nm sup 2) to O(m). Run time complexity is further reduced with parallel mapping algorithms based on these improvements, which run on the architecture for which they create the mappings.

The overall objective of this research is to develop efficient methods for the solution of linear and nonlinear systems of equations on parallel and supercomputers, and to apply these methods to the solution of problems in structural analysis. Attention has been given so far only to linear equations. The methods considered for the solution of the stiffness equation Kx=f have been Choleski factorization and the conjugate gradient iteration with SSOR and Incomplete Choleski preconditioning. More detail on these methods will be given on subsequent slides. These methods have been used to solve for the static displacements for the mast and panel focus problems in conjunction with the CSM testbed system based on NICE/SPAR.

Parallel finite element algorithms based on object-oriented concepts are presented. Moreover, the design and implementation of a data structure proposed are utilized in realizing a parallel geometric multigrid method. The ParFEMapper and the ParFECommunicator are the key components of the data structure in the proposed parallel scheme. These classes are constructed based on the type of finite elements (continuous or nonconforming or discontinuous) used. The proposed solver is compared with the open source direct solvers, MUMPS and PasTiX. Further, the performance of the parallel multigrid solver is analyzed up to 1080 processors. The solver shows a very good speedup up to 960 processors and the problem size has to be increased in order to maintain the good speedup when the number of processors are increased further. As a result, the parallel solver is able to handle large scale problems on massively parallel supercomputers. The proposed parallel finite element algorithms and multigrid solver are implemented in our in-house package ParMooN.

In Moving Horizon Estimation (MHE) the computed estimate is found by solving a constrained finite-time optimal estimation problem in real-time at each sample in a receding horizon fashion. The constrained estimation problem can be solved by, e.g., interior-point (IP) or active-set (AS) methods, where the main computational effort in both methods is known to be the computation of the search direction, i.e., the Newton step. This is often done using generic sparsity exploiting algorithms or serial Riccati recursions, but as parallel hardware is becoming more commonly available the need for parallel algorithms for computing the Newton step is increasing. In this paper a tailored, non-iterative parallel algorithm for computing the Newton step using the Riccati recursion is presented. The algorithm exploits the special structure of the Karush-Kuhn-Tucker system for the optimal estimation problem. As a result it is possible to obtain logarithmic complexity growth in the estimation horizon length, which can be used to reduce the computation time for IP and AS methods when applied to what is today considered as challenging estimation problems. Promising numerical results have been obtained using an ANSI-C implementation of the proposed algorithm running on true parallel hardware. Beyond MHE, due to similarities in the problem structure, the algorithm can be applied to various forms of on-line and off-line smoothing problems.

The objective of this research is to develop computationally efficient methods for solving fluid-structural interaction problems by directly coupling finite difference Euler/Navier-Stokes equations for fluids and finite element dynamics equations for structures on parallel computers. This capability will significantly impact many aerospace projects of national importance such as Advanced Subsonic Civil Transport (ASCT), where the structural stability margin becomes very critical at the transonic region. This research effort will have direct impact on the High Performance Computing and Communication (HPCC) Program of NASA in the area of parallel computing.

Implicit Runge-Kutta methods which are well-suited for parallel computations are characterized. It is claimed that such methods are first of all, those for which the associated rational approximation to the exponential has distinct poles, and these are called multiply explicit (MIRK) methods. Also, because of the so-called order reduction phenomenon, there is reason to require that these poles be real. Then, it is proved that a necessary condition for a q-stage, real MIRK to be A sub 0-stable with maximal order q + 1 is that q = 1, 2, 3, or 5. Nevertheless, it is shown that for every positive integer q, there exists a q-stage, real MIRK which is I-stable with order q. Finally, some useful examples of algebraically stable MIRKs are given.

The increasingly high relative cost of moving data on modern parallel machines has caused a paradigm shift in the design of high-performance algorithms: to achieve e ciency, one must focus on strategies which minimize data movement, rather than minimize arithmetic operations. We call this a communication-avoiding approach to algorithm design. In this work, we derive a new parallel communication-avoiding matrix powers algorithm for matrices of the form A = D+USV(H), where D is sparse and USV(H) has low rank but may be dense. Matrices of this form arise in many practical applications, including power-law graph analysis, circuit simulation and algorithms involving hierarchical (H) matrices, such as multigrid methods, fast multipole methods numerical partial di erential equation solvers, and preconditioned iterative methods. If A has this form, our algorithm enables a communication-avoiding approach. We demonstrate that, with respect to the cost of computing k sparse matrix-vector multiplications, our algorithm asymptotically reduces the parallel latency by a factor of O(k) for small additional bandwidth and computation costs. Using problems from real-world applications, our performance model predicts that this reduction in communication allows for up to 24 speedups on petascale machines.

Traditional parallel schedulers running on cluster supercomputers support only static scheduling, where the number of processors allocated to an application remains fixed throughout the execution of the job. This results in under-utilization of idle system resources thereby decreasing overall system throughput. In our research, we have developed a prototype framework called ReSHAPE, which supports dynamic resizing of parallel MPI applications executing on distributed memory platforms. The resizing library in ReSHAPE includes support for releasing and acquiring processors and efficiently redistributing application state to a new set of processors. In this paper, we derive an algorithm for redistributing two-dimensional block-cyclic arrays from $P$ to $Q$ processors, organized as 2-D processor grids. The algorithm ensures a contention-free communication schedule for data redistribution if $P_r \leq Q_r$ and $P_c \leq Q_c$. In other cases, the algorithm implements circular row and column shifts on the communication schedule to minimize node contention.

We present a multithreaded model for the dynamic load-balancing of numerical, adaptive computations required for the solution of Partial Differential Equations (PDEs) on multiprocessors. Multithreading is used as a means of exploring concurrency in the processor level in order to tolerate synchronization costs inherent to traditional (non-threaded) parallel adaptive PDE solvers. Our preliminary analysis for parallel, adaptive PDE solvers indicates that multithreading can be used as a mechanism to mask overheads required for the dynamic balancing of processor workloads with computations required for the actual numerical solution of the PDEs. Also, multithreading can simplify the implementation of dynamic load-balancing algorithms, a task that is very difficult for traditional data parallel adaptive PDE computations. Unfortunately, multithreading does not always simplify program complexity, often makes code re-usability not an easy task, and increases software complexity.

The primary objective of this project is to develop an advanced algorithm for parallel supercomputers to model time-dependent magnetohydrodynamics (MHD) in all three dimensions. This will provide a valuable tool for the design and testing of plasma related technologies that are important to the Air Force and industry. These applications include nuclear weapons effects simulations, radiation production for counter proliferation, fusion for power generation, and advanced plasma thrusters for space propulsion.

The objective of this work is to develop methodologies for the efficient computation of aerodynamic sensitivities using state-of-the-art CFD techniques for use in aerodynamic shape optimization and fluids-structure interaction analysis including: analysis of multiresolution schemes to compute sensitivities, analysis of parallel implementation issues in sensitivity computations, modification of a production CFD code for parallel multiresolution computation of sensitivities, integration of sensitivities into airframe optimization.

ParaDiGM the Parallel Distributed computation Graph Model, was designed to model implementations of parallel computations to be run on distributed message-based computer systems. We have used it to model two implementations of a complex adaptive parallel global optimization algorithm. In this paper, we introduce the ParaDiGM constructs, describe the algorithm, and then present the models of the implementations. These examples illustrate ParaDiGM's utility as a modeling formalism for representing and studying implementations of parallel, distributed algorithms.

This paper derives tradeoffs between three basic costs of a parallel algorithm: synchronization, data movement, and computational cost. Our theoretical model counts the amount of work and data movement as a maximum of any execution path during the parallel computation. By considering this metric, rather than the total communication volume over the whole machine, we obtain new insight into the characteristics of parallel schedules for algorithms with non-trivial dependency structures. The tradeoffs we derive are lower bounds on the execution time of the algorithm which are independent of the number of processors, but dependent on the problem size. Therefore, these tradeoffs provide lower bounds on the parallel execution time of any algorithm computed by a system composed of any number of homogeneous components each with associated computational, communication, and synchronization payloads. We first state our results for general graphs, based on expansion parameters, then we apply the theorem to a number of specific algorithms in numerical linear algebra, namely triangular substitution, Gaussian elimination, and Krylov subspace methods. Our lower bound for LU factorization demonstrates the optimality of Tiskin's LU algorithm [24] answering an open question posed in his paper, as well as of the 2.5D LU [20] algorithm which has analogous costs. We treat the computations in a general manner by noting that the computations share a similar dependency hypergraph structure and analyzing the communication requirements of lattice hypergraph structures.

The primary objective of this project is to develop an advanced algorithm for parallel supercomputers to model time-dependent magnetohydrodynamics (MHD) in all three dimensions. This will provide a valuable tool for the design and testing of plasma related technologies that are important to the Air Force and industry. Implementing the algorithm on parallel supercomputers will allow the detailed modeling of realistic plasmas in complex three-dimensional geometries. We have developed a time-dependent, two-dimensional, arbitrary-geometry version of the algorithm, placed it into a testbed code, added the modifications necessary for viscous and resistive effects, and tested the code against known analytical problems. We have implemented the algorithm on a parallel architecture and investigated parallelization strategies. Future plans include installing the algorithm into MACH2, optimizing the parallelization, extending the code to three dimensions, installing the three-dimensional algorithm into MACH3, and calibrating the code with experimental data.

We study the computational, communication, and scalability characteristics of a computational fluid dynamics application, which solves the time accurate flow field of a jet using the compressible Navier-Stokes equations, on a variety of parallel architecture platforms. The platforms chosen for this study are a cluster of workstations (the LACE experimental testbed at NASA Lewis), a shared memory multiprocessor (the Cray YMP), and distributed memory multiprocessors with different topologies - the IBM SP and the Cray T3D. We investigate the impact of various networks connecting the cluster of workstations on the performance of the application and the overheads induced by popular message passing libraries used for parallelization. The work also highlights the importance of matching the memory bandwidth to the processor speed for good single processor performance. By studying the performance of an application on a variety of architectures, we are able to point out the strengths and weaknesses of each of the example computing platforms.

Precise calculation of molecular electronic wavefunctions by methods such as coupled-cluster requires the computation of tensor contractions, the cost of which has polynomial computational scaling with respect to the system and basis set sizes. Each contraction may be executed via matrix multiplication on a properly ordered and structured tensor. However, data transpositions are often needed to reorder the tensors for each contraction. Writing and optimizing distributed-memory kernels for each transposition and contraction is tedious since the number of contractions scales combinatorially with the number of tensor indices. We present a distributed-memory numerical library (Cyclops Tensor Framework (CTF)) that automatically manages tensor blocking and redistribution to perform any user-specified contractions. CTF serves as the distributed memory contraction engine in Aquarius, a new program designed for high-accuracy and massively-parallel quantum chemical computations. Aquarius implements a range of coupled-cluster and related methods such as CCSD and CCSDT by writing the equations on top of a C++ templated domain-specific language. This DSL calls CTF directly to manage the data and perform the contractions. Our CCSD and CCSDT implementations achieve high parallel scalability on the BlueGene/Q and Cray XC30 supercomputer architectures showing that accurate electronic structure calculations can be effectively carried out on top of general distributed memory tensor primitives.

Explicit codes are often used to simulate the nonlinear dynamics of large-scale structural systems, even for low frequency response, because the storage and CPU requirements entailed by the repeated factorizations traditionally found in implicit codes rapidly overwhelm the available computing resources. With the advent of parallel processing, this trend is accelerating because explicit schemes are also easier to parallelize than implicit ones. However, the time step restriction imposed by the Courant stability condition on all explicit schemes cannot yet -- and perhaps will never -- be offset by the speed of parallel hardware. Therefore, it is essential to develop efficient and robust alternatives to direct methods that are also amenable to massively parallel processing because implicit codes using unconditionally stable time-integration algorithms are computationally more efficient when simulating low-frequency dynamics. Here we present a domain decomposition method for implicit schemes that requires significantly less storage than factorization algorithms, that is several times faster than other popular direct and iterative methods, that can be easily implemented on both shared and local memory parallel processors, and that is both computationally and communication-wise efficient. The proposed transient domain decomposition method is an extension of the method of Finite Element Tearing and Interconnecting (FETI) developed by Farhat and Roux for the solution of static problems. Serial and parallel performance results on the CRAY Y-MP/8 and the iPSC-860/128 systems are reported and analyzed for realistic structural dynamics problems. These results establish the superiority of the FETI method over both the serial/parallel conjugate gradient algorithm with diagonal scaling and the serial/parallel direct method, and contrast the computational power of the iPSC-860/128 parallel processor with that of the CRAY Y-MP/8 system.

A procedure for computing the aeroelasticity of wings on parallel multiple-instruction, multiple-data (MIMD) computers is presented. In this procedure, fluids are modeled using Euler equations, and structures are modeled using modal or finite element equations. The procedure is designed in such a way that each discipline can be developed and maintained independently by using a domain decomposition approach. In the present parallel procedure, each computational domain is scalable. A parallel integration scheme is used to compute aeroelastic responses by solving fluid and structural equations concurrently. The computational efficiency issues of parallel integration of both fluid and structural equations are investigated in detail. This approach, which reduces the total computational time by a factor of almost 2, is demonstrated for a typical aeroelastic wing by using various numbers of processors on the Intel iPSC/860.

The scope of this project dealt with the investigation of the requirements to support distributed computing of scientific computations over a cluster of cooperative workstations. Various experiments on computations for the solution of simultaneous linear equations were performed in the early phase of the project to gain experience in the general nature and requirements of scientific applications. A specification of a distributed integrated computing environment, DICE, based on a distributed shared memory communication paradigm has been developed and evaluated. The distributed shared memory model facilitates porting existing parallel algorithms that have been designed for shared memory multiprocessor systems to the new environment. The potential of this new environment is to provide supercomputing capability through the utilization of the aggregate power of workstations cooperating in a cluster interconnected via a local area network. Workstations, generally, do not have the computing power to tackle complex scientific applications, making them primarily useful for visualization, data reduction, and filtering as far as complex scientific applications are concerned. There is a tremendous amount of computing power that is left unused in a network of workstations. Very often a workstation is simply sitting idle on a desk. A set of tools can be developed to take advantage of this potential computing power to create a platform suitable for large scientific computations. The integration of several workstations into a logical cluster of distributed, cooperative, computing stations presents an alternative to shared memory multiprocessor systems. In this project we designed and evaluated such a system.

The effectiveness of any given mapping of workload to processors in a parallel system is dependent on the stochastic behavior of the workload. Program behavior is often characterized by a sequence of phases, with phase changes occurring unpredictably. During a phase, the behavior is fairly stable, but may become quite different during the next phase. Thus a workload assignment generated for one phase may hinder performance during the next phase. We consider the problem of deciding whether to remap a paralled computation in the face of uncertainty in remapping's utility. Fundamentally, it is necessary to balance the expected remapping performance gain against the delay cost of remapping. This paper treats this problem formally by constructing a probabilistic model of a computation with at most two phases. We use stochastic dynamic programming to show that the remapping decision policy which minimizes the expected running time of the computation has an extremely simple structure: the optimal decision at any step is followed by comparing the probability of remapping gain against a threshold. This theoretical result stresses the importance of detecting a phase change, and assessing the possibility of gain from remapping. We also empirically study the sensitivity of optimal performance to imprecise decision threshold. Under a wide range of model parameter values, we find nearly optimal performance if remapping is chosen simply when the gain probability is high. These results strongly suggest that except in extreme cases, the remapping decision problem is essentially that of dynamically determining whether gain can be achieved by remapping after a phase change; precise quantification of the decision model parameters is not necessary.

A large class of computations are characterized by a sequence of phases, with phase changes occurring unpredictably. The decision problem was considered regarding the remapping of workload to processors in a parallel computation when the utility of remapping and the future behavior of the workload is uncertain, and phases exhibit stable execution requirements during a given phase, but requirements may change radically between phases. For these problems a workload assignment generated for one phase may hinder performance during the next phase. This problem is treated formally for a probabilistic model of computation with at most two phases. The fundamental problem of balancing the expected remapping performance gain against the delay cost was addressed. Stochastic dynamic programming is used to show that the remapping decision policy minimizing the expected running time of the computation has an extremely simple structure. Because the gain may not be predictable, the performance of a heuristic policy that does not require estimnation of the gain is examined. The heuristic method's feasibility is demonstrated by its use on an adaptive fluid dynamics code on a multiprocessor. The results suggest that except in extreme cases, the remapping decision problem is essentially that of dynamically determining whether gain can be achieved by remapping after a phase change. The results also suggest that this heuristic is applicable to computations with more than two phases.

A large class of computations are characterized by a sequence of phases, with phase changes occurring unpredictably. The decision problem was considered regarding the remapping of workload to processors in a parallel computation when the utility of remapping and the future behavior of the workload is uncertain, and phases exhibit stable execution requirements during a given phase, but requirements may change radically between phases. For these problems a workload assignment generated for one phase may hinder performance during the next phase. This problem is treated formally for a probabilistic model of computation with at most two phases. The fundamental problem of balancing the expected remapping performance gain against the delay cost was addressed. Stochastic dynamic programming is used to show that the remapping decision policy minimizing the expected running time of the computation has an extremely simple structure. Because the gain may not be predictable, the performance of a heuristic policy that does not require estimnation of the gain is examined. The heuristic method's feasibility is demonstrated by its use on an adaptive fluid dynamics code on a multiprocessor. The results suggest that except in extreme cases, the remapping decision problem is essentially that of dynamically determining whether gain can be achieved by remapping after a phase change. The results also suggest that this heuristic is applicable to computations with more than two phases.

The purpose of this study is to investigate the performance of the integral equation computations using numerical source field-panel method in a massively parallel processing (MPP) environment. A comparative study of computational performance of the MPP CM-5 computer and conventional Cray-YMP supercomputer for a three-dimensional flow problem is made. A serial FORTRAN code is converted into a parallel CM-FORTRAN code. Some performance results are obtained on CM-5 with 32, 62, 128 nodes along with those on Cray-YMP with a single processor. The comparison of the performance indicates that the parallel CM-FORTRAN code near or out-performs the equivalent serial FORTRAN code for some cases.

A large class of computational problems is characterized by frequent synchronization, and computational requirements which change as a function of time. When such a problem must be solved on a message passing multiprocessor machine, the combination of these characteristics lead to system performance which decreases in time. Performance can be improved with periodic redistribution of computational load; however, redistribution can exact a sometimes large delay cost. We study the issue of deciding when to invoke a global load remapping mechanism. Such a decision policy must effectively weigh the costs of remapping against the performance benefits. We treat this problem by constructing two analytic models which exhibit stochastically decreasing performance. One model is quite tractable; we are able to describe the optimal remapping algorithm, and the optimal decision policy governing when to invoke that algorithm. However, computational complexity prohibits the use of the optimal remapping decision policy. We then study the performance of a general remapping policy on both analytic models. This policy attempts to minimize a statistic W(n) which measures the system degradation (including the cost of remapping) per computation step over a period of n steps. We show that as a function of time, the expected value of W(n) has at most one minimum, and that when this minimum exists it defines the optimal fixed-interval remapping policy. Our decision policy appeals to this result by remapping when it estimates that W(n) is minimized. Our performance data suggests that this policy effectively finds the natural frequency of remapping. We also use the analytic models to express the relationship between performance and remapping cost, number of processors, and the computation's stochastic activity.

The primary objective of this project is to develop an advanced algorithm for parallel supercomputers to model time- dependent magnetohydrodynamics (MHD) in all three dimensions. A production code with this algorithm will provide a valuable tool for the design and testing of plasma related technologies that are important to the Air Force and industry. Implementing the algorithm on parallel supercomputers will allow the detailed modeling of realistic plasmas in complex three-dimensional geometries. The algorithm incorporates an approximate Riemann solver with explicit diffusive terms. The system of equations is solved using an implicit lower-upper symmetric-Gauss-Seidel (LUSGS) relaxation method. A two-dimensional version of the algorithm has been developed, placed into a testhed code, modified to include viscous and resistive effects, and tested against known analytical problems. The algorithm has been benchmarked to the one-dimensional shock tube and two-dimensional fully-developed (unmagnetized) viscous flow and (magnetized) Hattmann flow. The algorithm has been implemented on a parallel architecture and parallelization strategies have been investigated.

This paper describes the algorithms and functionality of a new module developed to support overset grid assembly associated with performing time-dependent and adaptive moving body calculations of external aerodynamic flows using a multi-solver paradigm (i.e. different CFD solvers in different parts of the computational domain). We use the term domain connectivity in this paper to denote all the procedures that are involved in an overset grid assembly, and the module developed is referred henceforth as the domain-connectivity module. The domain-connectivity module coordinates the data transfer between different solvers applied in different parts of the computational domain -- body fitted structured or unstructured to capture viscous near-wall effects, and Cartesian adaptive mesh refinement to capture effects away from the wall. The execution of the CFD solvers and the domain-connectivity module are orchestrated by a Python-based computational infrastructure. The domain-connectivity module is fully parallel and performs all its operations (identification of grid overlaps and determination of data interpolation strategy) on the partitioned grid data. In addition, the domain connectivity procedures are completely automated such that no user intervention or manual input is necessary. The capabilities and performance of the package are presented for several test problems, including flow over a NACA 0015 wing and an AGARD A2 slotted airfoil, hover simulation of a scaled V-22 rotor, and dynamic simulation of a UH-60A rotor in forward flight. A modification to the algorithm for improved domain connectivity solutions in problems with tight tolerances as well as heterogeneous grid clustering is also presented.

The development and testing of a parallel unstructured agglomeration multigrid algorithm for steady-state aerodynamic flows is discussed. The agglomeration multigrid strategy uses a graph algorithm to construct the coarse multigrid levels from the given fine grid, similar to an algebraic multigrid approach, but operates directly on the non-linear system using the FAS (Full Approximation Scheme) approach. The scalability and convergence rate of the multigrid algorithm are examined on the SGI Origin 2000 and the Cray T3E. An argument is given which indicates that the asymptotic scalability of the multigrid algorithm should be similar to that of its underlying single grid smoothing scheme. For medium size problems involving several million grid points, near perfect scalability is obtained for the single grid algorithm, while only a slight drop-off in parallel efficiency is observed for the multigrid V- and W-cycles, using up to 128 processors on the SGI Origin 2000, and up to 512 processors on the Cray T3E. For a large problem using 25 million grid points, good scalability is observed for the multigrid algorithm using up to 1450 processors on a Cray T3E, even when the coarsest grid level contains fewer points than the total number of processors.

A hybrid scheme that utilizes MPI for distributed memory parallelism and OpenMP for shared memory parallelism is presented. The work is motivated by the desire to achieve exceptionally high Reynolds numbers in pseudospectral computations of fluid turbulence on emerging petascale, high core-count, massively parallel processing systems. The hybrid implementation derives from and augments a well-tested scalable MPI-parallelized pseudospectral code. The hybrid paradigm leads to a new picture for the domain decomposition of the pseudospectral grids, which is helpful in understanding, among other things, the 3D transpose of the global data that is necessary for the parallel fast Fourier transforms that are the central component of the numerical discretizations. Details of the hybrid implementation are provided, and performance tests illustrate the utility of the method. It is shown that the hybrid scheme achieves near ideal scalability up to ~20000 compute cores with a maximum mean efficiency of 83%. Data are presented that demonstrate how to choose the optimal number of MPI processes and OpenMP threads in order to optimize code performance on two different platforms.

Even though in recent years the scale of statistical analysis problems has increased tremendously, many statistical software tools are still limited to single-node computations. However, statistical analyses are largely based on dense linear algebra operations, which have been deeply studied, optimized and parallelized in the high-performance-computing community. To make high-performance distributed computations available for statistical analysis, and thus enable large scale statistical computations, we introduce RElem, an open source package that integrates the distributed dense linear algebra library Elemental into R. While on the one hand, RElem provides direct wrappers of Elemental's routines, on the other hand, it overloads various operators and functions to provide an entirely native R experience for distributed computations. We showcase how simple it is to port existing R programs to Relem and demonstrate that Relem indeed allows to scale beyond the single-node limitation of R with the full performance of Elemental without any overhead.

Even though in recent years the scale of statistical analysis problems has increased tremendously, many statistical software tools are still limited to single-node computations. However, statistical analyses are largely based on dense linear algebra operations, which have been deeply studied, optimized and parallelized in the high-performance-computing community. To make high-performance distributed computations available for statistical analysis, and thus enable large scale statistical computations, we introduce RElem, an open source package that integrates the distributed dense linear algebra library Elemental into R. While on the one hand, RElem provides direct wrappers of Elemental's routines, on the other hand, it overloads various operators and functions to provide an entirely native R experience for distributed computations. We showcase how simple it is to port existing R programs to Relem and demonstrate that Relem indeed allows to scale beyond the single-node limitation of R with the full performance of Elemental without any overhead.

A new algorithm for Monte Carlo calculation of the double exchange model is studied. The algorithm is commonly applicable to wide classes of strongly correlated electron systems which involve itinerant electrons coupled with thermodynamically fluctuating fields. Using moment expansions of the density of states with Chebyshev polynomials, the algorithm provides an efficient calculation on large size clusters, especially on parallel computers. Benchmark calculations are performed on Beowulf-type cluster systems with over 100 CPUs in parallel.

We introduce a high-performance virtual machine (VM) written in a numerically fast language like Fortran or C to evaluate very large expressions. We discuss the general concept of how to perform computations in terms of a VM and present specifically a VM that is able to compute tree-level cross sections for any number of external legs, given the corresponding byte code from the optimal matrix element generator, O'Mega. Furthermore, this approach allows to formulate the parallel computation of a single phase space point in a simple and obvious way. We analyze hereby the scaling behaviour with multiple threads as well as the benefits and drawbacks that are introduced with this method. Our implementation of a VM can run faster than the corresponding native, compiled code for certain processes and compilers, especially for very high multiplicities, and has in general runtimes in the same order of magnitude. By avoiding the tedious compile and link steps, which may fail for source code files of gigabyte sizes, new processes or complex higher order corrections that are currently out of reach could be evaluated with a VM given enough computing power.

This paper describes the parallelization strategy and achieved parallel efficiency of an explicit time-marching algorithm for solving conservation laws. The Space-Time Conservation Element and Solution Element (CE/SE) algorithm for solving the 2D and 3D Euler equations is parallelized with the aid of domain decomposition. The parallel efficiency of the resultant algorithm on a Silicon Graphics Origin 2000 parallel computer is checked.

Massively parallel simulations of transport equation systems call for a paradigm change in algorithm development to achieve efficient scalability. Traditional approaches require time synchronization of processing elements (PEs) which severely restricts scalability. Relaxing synchronization requirement introduces error and slows down convergence. In this paper, we propose and develop a novel `proxy-equation' concept for a general transport equation that (i) tolerates asynchrony with manageable added error, (ii) preserves convergence order and (iii) scales efficiently on massively parallel machines. The central idea is to modify \textit{a priori} the transport equation at the PE boundaries to offset asynchrony errors. Proof-of-concept computations are performed using a one-dimensional advection-diffusion equation. The results demonstrate the promise and advantages of the present strategy.

This is a survey paper on applications of mathematics of semirings to numerical analysis and computing. Concepts of universal algorithm and generic program are discussed. Relations between these concepts and mathematics of semirings are examined. A very brief introduction to mathematics of semirings (including idempotent and tropical mathematics) is presented. Concrete applications to optimization problems, idempotent linear algebra and interval analysis are indicated. It is known that some nonlinear problems (and especially optimization problems) become linear over appropriate semirings with idempotent addition (the so-called idempotent superposition principle). This linearity over semirings is convenient for parallel computations.