Wendell Anderson

THE INERTIAL MECHANISM IN THE MECHANICAL FILTRATION OF AEROSOLS

Abstract It has been shown that the inertial mechanism definitely exists and plays an important role in the mechanical filtration of aerosols at lower particle radii than calculated by Albrecht and Langmuir. The manner in which the inertial effect enters into the filtration mechanism can be expressed by the equations K p d e = 2rpv p 9n , K m d e = 2r 2 pv m 9n , where Kp and Km are constants, de is the effective fiber diameter, r is the particle radius, ρ is the particle density, vp is the velocity at which the inertial effect becomes measurable (penetration starts to decrease with increasing velocity), vm is the velocity above which no additional inertial effect occurs, and η is the viscosity of the gaseous medium. Tentative value for the constants Kp and Km have been determined experimentally. Calculated values of de for various filters are shown to be either direct measures of the actual fiber diameter (df), the interfiber distance (di), or some function of the two depending on the method of formation of the filter. The general structure of a complete three-dimensional filtration diagram for a given filter is indicated. The inertial effect is proposed as the reason that no maximum occurs in the penetration-radius curve in the predicted size range.

Early Experience with Scientific Programs on the Cray MTA-2

We describe our experiences porting and tuning three scientific programs to the Cray MTA-2, paying particular attention to the problems posed by I/O. We have measured the performance of each of the programs over many different machine configurations and we report on the scalability of each program. In addition, we compare the performance of the MTA with that of an SGI Origin running all three programs.

A computational steering system for studying microwave interactions with missile bodies

The paper describes a computer modeling and simulation system that supports computational steering, which is an effort to make the typical simulation workflow more efficient. Our system provides an interface that allows scientists to perform all of the steps in the simulation process in parallel and online. It uses a standard network flow visualization package, which has been extended to display graphical output in an immersive virtual environment such as a CAVE. Our system allows scientists to interactively manipulate simulation parameters and observe the results. It also supports inverse steering, where the user specifies the desired simulation result, and the system searches for the simulation parameters that achieve this result. Taken together, these capabilities allow scientists to more efficiently and effectively understand model behavior, as well as to search through simulation parameter space. The paper is also a case study of applying our system to the problem of simulating microwave interactions with missile bodies. Because these interactions are difficult to study experimentally, and have important effects on missile electronics, there is a strong desire to develop and validate simulation models of this phenomena.

Goal-orientated computational steering

Computational steering is a newly evolving paradigm for working with simulation models. It entails integration of model execution, observation and input data manipulation carried out concurrently in pursuit of rapid insight and goal achievement. Keys to effective computational steering include advanced visualization, high performance processing and intuitive user control. The Naval Research Laboratory (NRL) has been integrating facilities in its Virtual Reality Lab and High Performance Computing Center for application of computational steering to study effects of electromagnetic wave interactions using the HASP (High Accuracy Scattering and Propagation) modeling technique developed at NRL. We are also investigating automated inverse steering which involves incorporation of global optimization techniques to assist the user with tuning of parameter values to produce desired behaviors in complex models.

Early Experiences on the NRL Cray XD1

Scientists at the Naval Research Laboratory (NRL) are engaged in a broad spectrum of research. In order to provide the high performance computing resources to support that work NRL has recently obtained a three cabinet XD1 with 432 Opteron 275 dual core CPUs, 144 Vertex II FPGAs, and 6 Virtex 4 FPGAs. This paper will examine the applicability of the XD1 to these scientific problems

Modeling pulse propagation and scattering in a dispersive medium: performance of MPI/OpenMP hybrid code

Accurate modeling of pulse propagation and scattering is of great importance to the Navy. In a non-dispersive medium a fourth order in time and space 2-D finite difference time domain (FDTD) scheme representation of the linear wave equation can be used. However when the medium is dispersive one is required to take into account the frequency dependent attenuation and phase velocity. Using a theory first proposed by Blackstock, the linear wave equation has been modified by adding an additional term (the derivative of the convolution between the causal time domain propagation factor and the acoustic pressure) that takes into account the dispersive nature of the medium. This additional term transforms the calculation from one suitable to a workstation into one very much suited to a large-scale computational platform, both in terms of computation and memory. With appropriate distribution of data, good scaling can be achieved up to thousands of processors. Due to the simple structure of the code, it is easily parallelized using three different techniques: pure MPI, pure OpenMP and a hybrid MPI/OpenMP. We use this real life application to evaluate the performance of the latest multi-cpu/multicore platforms available from the DoD HPCMP

Spectral distortion in sampling rate conversion by zero-order polynomial interpolation

For such applications as digital beamforming in sonar and sampling rate conversion in digital audio systems, there is interest in fast, approximate interpolation techniques. Computationally, the simplest of these is to approximate the desired time series sample by the available sample which is nearest in time. This procedure is variously known as sample-and-hold interpolation, zero-order polynomial interpolation, or selective subsampling. The effect of this approximation on signal-to-noise ratio has already been addressed in the literature. This correspondence extends those previous analyses by deriving the spectral characteristics of the noise generated. It is shown that, because the timing errors involved in zero-order polynomial interpolation are not random, the effect on the input spectrum is not simply to broaden it by the transfer of power from narrowband signal components to broadband noise, from narrowband signals to narrowband noise or distortion components. >

Modeling Pulse Propagation and Scattering in a Dispersive Medium

Accurate modeling of pulse propagation and scattering is of great importance to the Navy. In a nondispersive medium, a fourth order in time and space 2-D Finite Difference Time Domain (FDTD) scheme representation of the linear wave equation can be used. However, when the medium is dispersive, one is required to take into account the frequency-dependent attenuation and phase velocity. Until recently, to include the dispersive effects, one typically solved the problem in the frequency domain and not in the time domain. The frequency domain solutions were Fourier transformed into the time domain. However by using a theory first proposed by Blackstock, the linear wave equation has been modified by adding an additional term (the derivative of the convolution between the causal time domain propagation factor and the acoustic pressure) that takes into account the dispersive nature of the medium. In the cases we are examining, what makes the water environment dispersive is the presence of air bubbles that are present at only a very small set of grid points. This leads to a numerical scheme where the amount of work to be done on individual grid points can vary by two orders of magnitude, leading to load balancing problems. This paper will examine the challenges encountered in implementing this problem efficiently on High Performance Computer systems and the tools, algorithms, and code changes required to run this code efficiently on such machines.

A Computational Steering System for Studying Microwave Interactions with Space-Borne Bodies

This paper describes a computer modeling and simulation systemthat supports computational steering, which is an effort to makethe typical simulation workflow more efficient. Our system providesan interface that allows scientists to perform all of the stepsin the simulation process in parallel and online. It uses a standardnetwork flow visualization package, which has been extended todisplay graphical output in an immersive virtual environment suchas a CAVE. Our system allows scientists to interactively manipulatesimulation parameters and observe the results. It also supportsinverse steering, where the user specifies the desired simulationresult, and the system searches for the simulation parameters thatachieve this result. Taken together, these capabilities allow scientiststo more efficiently and effectively understand model behavior,as well as to search through simulation parameter space.This paper is also a case study of applying our system to theproblem of simulating microwave interactions with missile bodies.Because these interactions are difficult to study experimentally, andhave important effects on missile electronics, there is a strong desireto develop and validate simulation models of this phenomena.

Evaluating High Performance Computing Systems at the Naval Research Laboratory

A leading edge center within the high performance computing modernization program (HPCMP), the Center for Computational Science (CCS) of the Naval Research Laboratory evaluates new high performance computing (HPC) assets. The center is currently evaluating two systems - an Altix 3700 running the SUSE Linux Enterprise Server (SLES) 10 operating systems and a Cray XD1. With respect to the Altix system, Naval Research Laboratory is looking at applications requiring large memory and testing the new Intel 10.0 compilers. For theXD1, we are examining the applicability of dual core processors and field programmable gate arrays (FPGAs) to HPC applications.