published_or_final_version / Computer Science / Doctoral / Doctor of Philosophy

Functional parallel programming techniques for feed-forward artificial neural networks trained using backpropagation learning are analyzed. In particular, the Data Parallel Haskell extension to the Glasgow Haskell Compiler is considered as a tool for achieving data parallelism. We find much potential and elegance in this method, and determine that a sufficiently large workload is critical in achieving real gains. Several additional features are recommended to increase usability and improve results on small datasets. / text

Signal Track Analytical Research Tool (START), is a parallel system for analyzing large-scale genomic data. Currently, genomic data analyses are usually performed by using custom scripts developed by individual research groups, and/or by the integrated use of multiple existing tools (such as BEDTools and Galaxy). The goals of START are 1) to provide a single tool that supports a wide spectrum of genomic data analyses that are commonly done by analysts; and 2) to greatly simplify these analysis tasks by means of a simple declarative language (STQL) with which users only need to specify what they want to do, rather than the detailed computational steps as to how the analysis task should be performed. START consists of four major components: 1) A declarative language called Signal Track Query Language (STQL), which is a SQL-like language we specifically designed to suit the needs for analyzing genomic signal tracks. 2) A STQL processing system built on top of a large-scale distributed architecture. The system is based on the Hadoop distributed storage and the MapReduce Big Data processing framework. It processes each user query using multiple machines in parallel. 3) A simple and user-friendly web site that helps users construct and execute queries, upload/download compressed data files in various formats, man-age stored data, queries and analysis results, and share queries with other users. It also provides a complete help system, detailed specification of STQL, and a large number of sample queries for users to learn STQL and try START easily. Private files and queries are not accessible by other users. 4) A repository of public data popularly used for large-scale genomic data analysis, including data from ENCODE and Roadmap Epigenomics, that users can use in their analyses. / published_or_final_version / Computer Science / Doctoral / Doctor of Philosophy

As people are becoming increasingly dependent on computerized systems, the need for these systems to be dependable is also increasing. However, programming dependable systems is difficult, especially when parallelism is involved. This is due in part to the fact that very few high-level programming languages support both fault-tolerance and parallel programming. This dissertation addresses this problem by presenting FT-Linda, a high-level language for programming fault-tolerant parallel programs. FT-Linda is based on Linda, a language for programming parallel applications whose most notable feature is a distributed shared memory called tuple space. FT-Linda extends Linda by providing support to allow a program to tolerate failures in the underlying computing platform. The distinguishing features of FT-Linda are stable tuple spaces and atomic execution of multiple tuple space operations. The former is a type of stable storage in which tuple values are guaranteed to persist across failures, while the latter allows collections of tuple operations to be executed in an all-or-nothing fashion despite failures and concurrency. Example FT-Linda programs are given for both dependable systems and parallel applications. The design and implementation of FT-Linda are presented in detail. The key technique used is the replicated state machine approach to constructing fault-tolerant distributed programs. Here, tuple space is replicated to provide failure resilience, and the replicas are sent a message describing the atomic sequence of tuple space operations to perform. This strategy allows an efficient implementation in which only a single multicast message is needed for each atomic sequence of tuple space operations. An implementation of FT-Linda for a network of workstations is also described. FT-Linda is being implemented using Consul, a communication substrate that supports fault-tolerant distributed programming. Consul is built in turn with the x-kernel, an operating system kernel that provides support for composing network protocols. Each of the components of the implementation has been built and tested.

No description available.

With a modest beginning from developing a model of dynamics of hard liquid spheres (Alder et al., 1957), molecular dynamics (MD) simulations have come to a point where complex biomolecules can be simulated with precision close to reality (Noskov et al., 2007). In this context, a parallel molecular dynamics program for simulation of ion channels associated with cellular membranes has been developed. The parallel MD code developed is simple, efficient, and easily coupled to other codes such as the hybrid molecular dynamics/ brownian dynamics (MD/BD) code developed for the study of protein interactions (Ying et al., 2005). The Atom Decomposition (AD) Method was used in partitioning calculations on atoms to processors. One of the major impediments in using AD was the relatively large size of data that had to be communicated by the processes (Plimpton et al., 1995). Replicating only positions of atoms eased the congestion created by communication of both force terms and positions of atoms between processes. The performance of the code was tested on KcsA, a bacterial potassium channel. The program was written in Fortran 90 with parallel functions from the library of mpich-1.2.7. The idle time of processes was optimized by message driven ordering of communication. The scaling of the parallel program with 2000 – 60,000 atoms was determined and compared with the results obtained from the serial program. As expected, the parallel program scaled better than the serial program as the number of atoms included in the simulation increased from 2000 - 60000. The performance of the parallel program was tested on 4-15 processes, for a system comprising 20,000 atoms. The results obtained were compared with results from the serial program. It was observed that the parallel program scaled better than the serial program as the number of processes increased from 4 to 15. When compared with serial program, which had application of Newton’s Third Law in calculating force terms once per each pair of atoms, it was observed that the parallel program scaled better on 6-15 processes for a physical system comprising of 20,000 atoms.

The domain of nanoscale optical science and technology is a combination of the classical world of electromagnetics and the quantum mechanical regime of atoms and molecules. Recent advancements in fabrication technology allows the optical structures to be scaled down to nanoscale size or even to the atomic level, which are far smaller than the wavelength they are designed for. These nanostructures can have unique, controllable, and tunable optical properties and their interactions with quantum materials can have important near-field and far-field optical response. Undoubtedly, these optical properties can have many important applications, ranging from the efficient and tunable light sources, detectors, filters, modulators, high-speed all-optical switches; to the next-generation classical and quantum computation, and biophotonic medical sensors. This emerging research of nanoscience, known as nanophotonics, is a highly interdisciplinary field requiring expertise in materials science, physics, electrical engineering, and scientific computing, modeling and simulation. It has also become an important research field for investigating the science and engineering of light-matter interactions that take place on wavelength and subwavelength scales where the nature of the nanostructured matter controls the interactions. In addition, the fast advancements in the computing capabilities, such as parallel computing, also become as a critical element for investigating advanced nanophotonic devices. This role has taken on even greater urgency with the scale-down of device dimensions, and the design for these devices require extensive memory and extremely long core hours. Thus distributed computing platforms associated with parallel computing are required for faster designs processes. Scientific parallel computing constructs mathematical models and quantitative analysis techniques, and uses the computing machines to analyze and solve otherwise intractable scientific challenges. In particular, parallel computing are forms of computation operating on the principle that large problems can often be divided into smaller ones, which are then solved concurrently. In this dissertation, we report a series of new nanophotonic developments using the advanced parallel computing techniques. The applications include the structure optimizations at the nanoscale to control both the electromagnetic response of materials, and to manipulate nanoscale structures for enhanced field concentration, which enable breakthroughs in imaging, sensing systems (chapter 3 and 4) and improve the spatial-temporal resolutions of spectroscopies (chapter 5). We also report the investigations on the confinement study of optical-matter interactions at the quantum mechanical regime, where the size-dependent novel properties enhanced a wide range of technologies from the tunable and efficient light sources, detectors, to other nanophotonic elements with enhanced functionality (chapter 6 and 7).

Parallel computing is essential for solving very large scientific and engineering problems. An effective parallel computing solution requires an appropriate parallel machine and a well-optimised parallel program, both of which can be selected via performance modelling. This dissertation describes a new performance modelling system, called the Performance Evaluating Virtual Parallel Machine (PEVPM). Unlike previous techniques, the PEVPM system is relatively easy to use, inexpensive to apply and extremely accurate. It uses a novel bottom-up approach, where submodels of individual computation and communication events are dynamically constructed from data-dependencies, current contention levels and the performance distributions of low-level operations, which define performance variability in the face of contention. During model evaluation, the performance distribution attached to each submodel is sampled using Monte Carlo techniques, thus simulating the effects of contention. This allows the PEVPM to accurately simulate a program's execution structure, even if it is non-deterministic, and thus to predict its performance. Obtaining these performance distributions required the development of a new benchmarking tool, called MPIBench. Unlike previous tools, which simply measure average message-passing time over a large number of repeated message transfers, MPIBench uses a highly accurate and globally synchronised clock to measure the performance of individual communication operations. MPIBench was used to benchmark three parallel computers, which encompassed a wide range of network performance capabilities, namely those provided by Fast Ethernet, Myrinet and QsNet. Network contention, a problem ignored by most research in this area, was found to cause extensive performance variation during message-passing operations. For point-to-point communication, this variation was best described by Pearson 5 distributions. Collective communication operations were able to be modelled using their constituent point-to-point operations. In cases of severe contention, extreme outliers were common in the observed performance distributions, which were shown to be the result of lost messages and their subsequent retransmit timeouts. The highly accurate benchmark results provided by MPIBench were coupled with the PEVPM models of a range of parallel programs, and simulated by the PEVPM. These case studies proved that, unlike previous modelling approaches, the PEVPM technique successfully unites generality, flexibility, cost-effectiveness and accuracy in one performance modelling system for parallel programs. This makes it avaluable tool for the development of parallel computing solutions. / Thesis (Ph.D.)--Computer Science, 2003.

Two prevalent models of parallel programming are data parallelism and task parallelism. Data parallelism is the simultaneous application of a single operation to a data set. This model fits best with regular computations. Task parallelism is the simultaneous application of possibly different operations to possibly different data sets. This fits best with irregular computations. Efficient solution of some problems require both regular and irregular computations. Implementing efficient and portable parallel solutions to these problems requires a high-level language that can accommodate both task and data parallelism. We have extended the data-parallel language Dataparallel C to include task parallelism so that programmers may now use data and task parallelism within the same program. The extension permits the nesting of data-parallel constructs inside a task-parallel framework. We present a banded linear system to analyze the benefits of our language extensions. / Graduation date: 1995

M.S. / Computer Science & Engineering / The tasks associated with designing and. implementing parallel programs involve effectively partitioning the problem, defining an efficient. control strategy and mapping the design to a particular system. The task then becomes one of analyzing the program for correctness and stepwise refinement of its performance. New tools are needed to assist the programmer with these last two stages. Metrics and methods of instrumentation are needed to help with behavior analysis (debugging) and performance analysis. First, current tools and analysis methods are reviewed, and then a set of models is proposed for analyzing parallel programs. The design of IPPM, based on these models, is then presented. IPPM is an interactive, parallel program monitor for the Intel iPSC. It gives a post-mortem view of an iPSC program based on a script of events collected during execution. A user can observe changes in program state and synchronization, select statistics, interactively filter events and time critical sequences.