Engineering topics
Publications and source records attributed to Kutler, Paul.
This paper discusses the efforts to create, using two NASA SP2 supercomputers, a 'Metacenter' which includes the capability to transparently and dynamically distribute the SP2 workload across the geographically separated systems. Functional components of the Phase 1 Metacenter are identified, outstanding issues are discussed, and the plan for the second phase of the project is outlined.
We evaluate the performance of a Fast Ethernet network configured with a single large switch, a single hub, and a 4x4 2D torus topology in a testbed cluster of "commodity" Pentium Pro PCs. We also evaluated a mixed network composed of ethernet hubs and switches. An MPI collective communication benchmark, and the NAS Parallel Benchmarks version 2.2 (NPB2) show that the torus network performs best for all sizes that we were able to test (up to 16 nodes). For larger networks the ethernet switch outperforms the hub, though its performance is far less than peak. The hub/switch combination tests indicate that the NAS parallel benchmarks are relatively insensitive to hub densities of less than 7 nodes per hub.
The Whitney project is integrating commodity off-the-shelf PC hardware and software technology to build a parallel supercomputer with hundreds to thousands of nodes. To build such a system, one must have a scalable software model, and the installation and maintenance of the system software must be completely automated. We describe the design of an architecture for booting, installing, and configuring nodes in such a system with particular consideration given to scalability and ease of maintenance. This system has been implemented on a 40-node prototype of Whitney and is to be used on the 500 processor Whitney system to be built in 1998.
Recent advances in low-end processor and network technology have made it possible to build a "supercomputer" out of commodity components. We develop simple models of the NAS Parallel Benchmarks version 2 (NPB 2) to explore the cost/performance trade-offs involved in building a balanced parallel computer supporting a scientific workload. We develop closed form expressions detailing the number and size of messages sent by each benchmark. Coupling these with measured single processor performance, network latency, and network bandwidth, our models predict benchmark performance to within 30%. A comparison based on total system cost reveals that current commodity technology (200 MHz Pentium Pros with 100baseT Ethernet) is well balanced for the NPBs up to a total system cost of around $1,000,000.
A variety of different network technologies and topologies are currently being evaluated as part of the Whitney Project. This paper reports on the implementation and performance of a Fast Ethernet network configured in a 4x4 2D torus topology in a testbed cluster of 'commodity' Pentium Pro PCs. Several benchmarks were used for performance evaluation: an MPI point to point message passing benchmark, an MPI collective communication benchmark, and the NAS Parallel Benchmarks version 2.2 (NPB2). Our results show that for point to point communication on an unloaded network, the hub and 1 hop routes on the torus have about the same bandwidth and latency. However, the bandwidth decreases and the latency increases on the torus for each additional route hop. Collective communication benchmarks show that the torus provides roughly four times more aggregate bandwidth and eight times faster MPI barrier synchronizations than a hub based network for 16 processor systems. Finally, the SOAPBOX benchmarks, which simulate real-world CFD applications, generally demonstrated substantially better performance on the torus than on the hub. In the few cases the hub was faster, the difference was negligible. In total, our experimental results lead to the conclusion that for Fast Ethernet networks, the torus topology has better performance and scales better than a hub based network.
In the area of scientific visualization, input data sets are often very large. In visualization of Computational Fluid Dynamics (CFD) in particular, input data sets today can surpass 100 Gbytes, and are expected to scale with the ability of supercomputers to generate them. Some visualization tools already partition large data sets into segments, and load appropriate segments as they are needed. However, this does not remove the problem for two reasons: 1) there are data sets for which even the individual segments are too large for the largest graphics workstations, 2) many practitioners do not have access to workstations with the memory capacity required to load even a segment, especially since the state-of-the-art visualization tools tend to be developed by researchers with much more powerful machines. When the size of the data that must be accessed is larger than the size of memory, some form of virtual memory is simply required. This may be by segmentation, paging, or by paged segments. In this paper we demonstrate that complete reliance on operating system virtual memory for out-of-core visualization leads to poor performance. We then describe a paged segment system that we have implemented, and explore the principles of memory management that can be employed by the application for out-of-core visualization. We show that application control over some of these can significantly improve performance. We show that sparse traversal can be exploited by loading only those data actually required. We show also that application control over data loading can be exploited by 1) loading data from alternative storage format (in particular 3-dimensional data stored in sub-cubes), 2) controlling the page size. Both of these techniques effectively reduce the total memory required by visualization at run-time. We also describe experiments we have done on remote out-of-core visualization (when pages are read by demand from remote disk) whose results are promising.
The emergence of rapidly expanding technologies for distribution and dissemination of information and knowledge has brought to focus the opportunities for development of knowledge-based networks, knowledge dissemination and knowledge management technologies and their potential applications for enhancing productivity of knowledge work. The challenging and complex problems of the future can be best addressed by developing the knowledge management as a new discipline based on an integrative synthesis of hard and soft sciences. A knowledge management professional society can provide a framework for catalyzing the development of proposed synthesis as well as serve as a focal point for coordination of professional activities in the strategic areas of education, research and technology development. Preliminary concepts for the development of the knowledge management discipline and the professional society are explored. Within this context of knowledge management discipline and the professional society, potential opportunities for application of information technologies for more effectively delivering or transferring information and knowledge (i.e., resulting from the NASA's Mission to Planet Earth) for the development of policy options in critical areas of national and global importance (i.e., policy decisions in economic and environmental areas) can be explored, particularly for those policy areas where a global collaborative knowledge network is likely to be critical to the acceptance of the policies.
NPB2 (NAS (NASA Advanced Supercomputing) Parallel Benchmarks 2) is an implementation, based on Fortran and the MPI (message passing interface) message passing standard, of the original NAS Parallel Benchmark specifications. NPB2 programs are run with little or no tuning, in contrast to NPB vendor implementations, which are highly optimized for specific architectures. NPB2 results complement, rather than replace, NPB results. Because they have not been optimized by vendors, NPB2 implementations approximate the performance a typical user can expect for a portable parallel program on distributed memory parallel computers. Together these results provide an insightful comparison of the real-world performance of high-performance computers. New NPB2 features: New implementation (CG), new workstation class problem sizes, new serial sample versions, more performance statistics.
The two commonly-used performance data types in the super-computing community, statistics and event traces, are discussed and compared. Statistical data are much more compact but lack the probative power event traces offer. Event traces, on the other hand, are unbounded and can easily fill up the entire file system during program execution. In this paper, we propose an innovative methodology for performance data gathering and representation that offers a middle ground. Two basic ideas are employed: the use of averages to replace recording data for each instance and 'formulae' to represent sequences associated with communication and control flow. The user can trade off tracing overhead, trace data size with data quality incrementally. In other words, the user will be able to limit the amount of trace data collected and, at the same time, carry out some of the analysis event traces offer using space-time views. With the help of a few simple examples, we illustrate the use of these techniques in performance tuning and compare the quality of the traces we collected with event traces. We found that the trace files thus obtained are, indeed, small, bounded and predictable before program execution, and that the quality of the space-time views generated from these statistical data are excellent. Furthermore, experimental results showed that the formulae proposed were able to capture all the sequences associated with 11 of the 15 applications tested. The performance of the formulae can be incrementally improved by allocating more memory at runtime to learn longer sequences.
Temperature effects can be important in short pulse propagation or in CW operation of VCSELs. In those cases, the optical susceptibility function is, in general, a function of carrier density and plasma and/or lattice temperature(s). Just as there is an alpha factor induced by the density change (density alpha), there are alpha factors induced by lattice or plasma temperature change (temperature alphas). The density alpha factor has been studied quite extensively, while the temperature alpha factors have not been computed for any structure. I will report on my first calculations of alpha factors induced by the plasma and lattice temperature change for GaAs/AlGaAs quantum wells using microscopic theory. Besides, the temperature dependence of the density alpha factor is also considered.
In several cases, new visualization techniques have vastly increased the researcher's ability to analyze and comprehend data. Similarly, the role of networks in providing an efficient supercomputing environment have become more critical and continue to grow at a faster rate than the increase in the processing capabilities of supercomputers. A close relationship between scientific visualization and high-speed networks in providing an important link to support efficient supercomputing is identified. The two technologies are driven by the increasing complexities and volume of supercomputer data. The interaction of scientific visualization and high-speed networks in a Computational Fluid Dynamics simulation/visualization environment are given. Current capabilities supported by high speed networks, supercomputers, and high-performance graphics workstations at the Numerical Aerodynamic Simulation Facility (NAS) at NASA Ames Research Center are described. Applied research in providing a supercomputer visualization environment to support future computational requirements are summarized.
The i860 microprocessor is a pipelined processor that can deliver two double precision floating point results every clock. It is being used in the Touchstone project to develop a teraflop computer by the year 2000. With such high computational capabilities it was expected that memory bandwidth would limit performance on many kernels. Measured performance of three kernels showed performance is less than what memory bandwidth limitations would predict. This paper develops a model that explains the discrepancy in terms of memory latencies and points to some problems involved in moving data from memory to the arithmetic pipelines.
In a previous report the design concepts of Charon were presented. Charon is a toolkit that aids engineers in developing scientific programs for structured-grid applications to be run on MIMD parallel computers. It constitutes an augmentation of the general-purpose MPI-based message-passing layer, and provides the user with a hierarchy of tools for rapid prototyping and validation of parallel programs, and subsequent piecemeal performance tuning. Here we describe the implementation of the domain decomposition tools used for creating data distributions across sets of processors. We also present the hierarchy of parallelization tools that allows smooth translation of legacy code (or a serial design) into a parallel program. Along with the actual tool descriptions, we will present the considerations that led to the particular design choices. Many of these are motivated by the requirement that Charon must be useful within the traditional computational environments of Fortran 77 and C. Only the Fortran 77 syntax will be presented in this report.
Modem aircraft design involves study of airflow through both windtunnel testing and computer simulation. These computer simulations result in often very large and complex sets of numbers, which contain information critical to the aircrafts performance. This talk will describe how visualization is used to understand these simulations, using a variety of techniques including low-level analysis such as simulated particles, high-level feature detection, and virtual-reality-based techniques for exploration. We will focus on the challenges of extremely large data sets, interactive performance, and information extraction. The talk will close with a vision of the future including the integration of simulation and visualization.
Due to the rapid progress in atom manipulation technology, atomic chain electronics would not be a dream, where foreign atoms are placed on a substrate to form a chain, and its electronic properties are designed by controlling the lattice constant d. It has been shown theoretically that a Si atomic chain is metallic regardless of d and that a Mg atomic chain is semiconducting or insulating with a band gap modified with d. For electronic applications, it is essential to establish a method to dope a semiconducting chain, which is to control the Fermi energy position without altering the original band structure. If we replace some of the chain atoms with dopant atoms randomly, the electrons will see random potential along the chain and will be localized strongly in space (Anderson localization). However, if we replace periodically, although the electrons can spread over the chain, there will generally appear new bands and band gaps reflecting the new periodicity of dopant atoms. This will change the original band structure significantly. In order to overcome this dilemma, we may place a dopant atom beside the chain at every N lattice periods (N > 1). Because of the periodic arrangement of dopant atoms, we can avoid the unwanted Anderson localization. Moreover, since the dopant atoms do not constitute the chain, the overlap interaction between them is minimized, and the band structure modification can be made smallest. Some tight-binding results will be discussed to demonstrate the present idea.
Concern is increasing about how quantum effects will impact electronic device operation as down-scaling continues along the SIA Roadmap through 2010. This document describes part of a new semiconductor device modeling (SDM) program at NAS to investigate these concerns by utilizing advanced NAS and third-party numerical computation software to rapidly implement and investigate electronic device models including quantum effects. This SDM project will investigate quantum effects in devices in the classical-quantum transition region, especially sub-0.1 mm MOSFETs. Specific tasks planned for this project include the use of quantum corrections to the classical drift-diffusion and hydrodynamic models of electron transport, arid the use of nominally quantum models including significant scattering.
The linear response conductance of a defect free carbon nanotube connected to ideal contacts is approximately 4e(sup 2)/h. In this paper, we first calculate the conductance characteristics of a uniform tube in the presence of an applied bias by assuring different forms of the potential drop across the wire. It is shown that for long tubes, the conduction mechanism involve a only the four non degenerate subands when the electric field drops slowly across the the tube. We then proceed to calculate the transmission and low bias conductance of tubes with defects. Here, we show that a weak uniform disorder in the tube does not cause a dramatic change in the conductance. However the presence of certain substitutional defects, the transmission coefficient versus energy develops a gap and this can cause a dramatic change in the low bias conductance. The effect of coherence is neglected.
This paper describes the performance characteristics of the computational workloads on the NAS Cray Y-MP machines, a Y-MP 832 and later a Y-MP 8128. Hardware measurements indicated that the Y-MP workload performance matured over time, ultimately sustaining an average throughput of 0.8 GFLOPS and a vector operation fraction of 87%. The measurements also revealed an operation rate exceeding 1 per clock period, a well-balanced architecture featuring a strong utilization of vector functional units, and an efficient memory organization. Introduction of the larger memory 8128 increased throughput by allowing a more efficient utilization of CPUs. Throughput also depended on the metering of the batch queues; low-idle Saturday workloads required a buffer of small jobs to prevent memory starvation of the CPU. UNICOS required about 7% of total CPU time to service the 832 workloads; this overhead decreased to 5% for the 8128 workloads. While most of the system time went to service I/O requests, efficient scheduling prevented excessive idle due to I/O wait. System measurements disclosed no obvious bottlenecks in the response of the machine and UNICOS to the workloads. In most cases, Cray-provided software tools were- quite sufficient for measuring the performance of both the machine and operating, system.