Quantifying ice-shelf basal melt and ice-stream dynamics using high-resolution DEM and GPS time series

Thesis (Ph.D.)--University of Washington, 2016-06

Submodular Optimization and Data Processing

Thesis (Ph.D.)--University of Washington, 2016-06

Neuro-electromagnetic imaging of the human visual cortex

Methods of solving the neuro-electromagnetic inverse problem are examined and developed, with specific reference to the human visual cortex. The anatomy, physiology and function of the human visual system are first reviewed. Mechanisms by which the visual cortex gives rise to external electric and magnetic fields are then discussed, and the forward problem is described mathematically for the case of an isotropic, piecewise homogeneous volume conductor, and then for an anisotropic, concentric, spherical volume conductor. Methods of solving the inverse problem are reviewed, before a new technique is presented. This technique combines prior anatomical information gained from stereotaxic studies, with a probabilistic distributed-source algorithm to yield accurate, realistic inverse solutions. The solution accuracy is enhanced by using both visual evoked electric and magnetic responses simultaneously. The numerical algorithm is then modified to perform equivalent current dipole fitting and minimum norm estimation, and these three techniques are implemented on a transputer array for fast computation. Due to the linear nature of the techniques, they can be executed on up to 22 transputers with close to linear speedup. The latter part of the thesis describes the application of the inverse methods to the analysis of visual evoked electric and magnetic responses. The CIIm peak of the pattern onset evoked magnetic response is deduced to be a product of current flowing away from the surface areas 17, 18 and 19, while the pattern reversal P100m response originates in the same areas, but from oppositely directed current. Cortical retinotopy is examined using sectorial stimuli, the CI and CIm ;peaks of the pattern onset electric and magnetic responses are found to originate from areas V1 and V2 simultaneously, and they therefore do not conform to a simple cruciform model of primary visual cortex.

Efficient asynchronous scheduling algorithms for cost-effective buffered crossbar switches

Buffered crossbar switches have recently attracted considerable attention as the next generation of high speed interconnects. They are a special type of crossbar switches with an exclusive buffer at each crosspoint of the crossbar. They demonstrate unique advantages over traditional unbuffered crossbar switches, such as high throughput, low latency, and asynchronous packet scheduling. However, since crosspoint buffers are expensive on-chip memories, it is desired that each crosspoint has only a small buffer. This dissertation proposes a series of practical algorithms and techniques for efficient packet scheduling for buffered crossbar switches. To reduce the hardware cost of such switches and make them scalable, we considered partially buffered crossbars, whose crosspoint buffers can be of an arbitrarily small size. Firstly, we introduced a hybrid scheme called Packet-mode Asynchronous Scheduling Algorithm (PASA) to schedule best effort traffic. PASA combines the features of both distributed and centralized scheduling algorithms and can directly handle variable length packets without Segmentation And Reassembly (SAR). We showed by theoretical analysis that it achieves 100% throughput for any admissible traffic in a crossbar with a speedup of two. Moreover, outputs in PASA have a large probability to avoid the more time-consuming centralized scheduling process, and thus make fast scheduling decisions. Secondly, we proposed the Fair Asynchronous Segment Scheduling (FASS) algorithm to handle guaranteed performance traffic with explicit flow rates. FASS reduces the crosspoint buffer size by dividing packets into shorter segments before transmission. It also provides tight constant performance guarantees by emulating the ideal Generalized Processor Sharing (GPS) model. Furthermore, FASS requires no speedup for the crossbar, lowering the hardware cost and improving the switch capacity. Thirdly, we presented a bandwidth allocation scheme called Queue Length Proportional (QLP) to apply FASS to best effort traffic. QLP dynamically obtains a feasible bandwidth allocation matrix based on the queue length information, and thus assists the crossbar switch to be more work-conserving. The feasibility and stability of QLP were proved, no matter whether the traffic distribution is uniform or non-uniform. Hence, based on bandwidth allocation of QLP, FASS can also achieve 100% throughput for best effort traffic in a crossbar without speedup.

Background traffic modeling for large-scale network simulation

Network simulation is an indispensable tool for studying Internet-scale networks due to the heterogeneous structure, immense size and changing properties. It is crucial for network simulators to generate representative traffic, which is necessary for effectively evaluating next-generation network protocols and applications. With network simulation, we can make a distinction between foreground traffic, which is generated by the target applications the researchers intend to study and therefore must be simulated with high fidelity, and background traffic, which represents the network traffic that is generated by other applications and does not require significant accuracy. The background traffic has a significant impact on the foreground traffic, since it competes with the foreground traffic for network resources and therefore can drastically affect the behavior of the applications that produce the foreground traffic. This dissertation aims to provide a solution to meaningfully generate background traffic in three aspects. First is realism. Realistic traffic characterization plays an important role in determining the correct outcome of the simulation studies. This work starts from enhancing an existing fluid background traffic model by removing its two unrealistic assumptions. The improved model can correctly reflect the network conditions in the reverse direction of the data traffic and can reproduce the traffic burstiness observed from measurements. Second is scalability. The trade-off between accuracy and scalability is a constant theme in background traffic modeling. This work presents a fast rate-based TCP (RTCP) traffic model, which originally used analytical models to represent TCP congestion control behavior. This model outperforms other existing traffic models in that it can correctly capture the overall TCP behavior and achieve a speedup of more than two orders of magnitude over the corresponding packet-oriented simulation. Third is network-wide traffic generation. Regardless of how detailed or scalable the models are, they mainly focus on how to generate traffic on one single link, which cannot be extended easily to studies of more complicated network scenarios. This work presents a cluster-based spatio-temporal background traffic generation model that considers spatial and temporal traffic characteristics as well as their correlations. The resulting model can be used effectively for the evaluation work in network studies.

Um algoritmo paralelo eficiente de migração reversa no tempo (rtm) 3d com granularidade fina

The reverse time migration algorithm (RTM) has been widely used in the seismic industry to generate images of the underground and thus reduce the risk of oil and gas exploration. Its widespread use is due to its high quality in underground imaging. The RTM is also known for its high computational cost. Therefore, parallel computing techniques have been used in their implementations. In general, parallel approaches for RTM use a coarse granularity by distributing the processing of a subset of seismic shots among nodes of distributed systems. Parallel approaches with coarse granularity for RTM have been shown to be very efficient since the processing of each seismic shot can be performed independently. For this reason, RTM algorithm performance can be considerably improved by using a parallel approach with finer granularity for the processing assigned to each node. This work presents an efficient parallel algorithm for 3D reverse time migration with fine granularity using OpenMP. The propagation algorithm of 3D acoustic wave makes up much of the RTM. Different load balancing were analyzed in order to minimize possible losses parallel performance at this stage. The results served as a basis for the implementation of other phases RTM: backpropagation and imaging condition. The proposed algorithm was tested with synthetic data representing some of the possible underground structures. Metrics such as speedup and efficiency were used to analyze its parallel performance. The migrated sections show that the algorithm obtained satisfactory performance in identifying subsurface structures. As for the parallel performance, the analysis clearly demonstrate the scalability of the algorithm achieving a speedup of 22.46 for the propagation of the wave and 16.95 for the RTM, both with 24 threads.

Aceleración de algoritmos de estimación de movimiento mediante OpenCL

El flujo óptico y la estimación de movimiento es área de conocimiento muy importante usado en otros campos del conocimiento como el de la seguridad o el de la bioinformática. En estos sectores, se demandan aplicaciones de flujo óptico que realicen actividades muy importantes con tiempos de ejecución lo más bajos posibles, llegando a tiempo real si es posible. Debido a la gran complejidad de cálculos que siguen a este tipo de algoritmos como se observará en la sección de resultados, la aceleración de estos es una parte vital para dar soporte y conseguir ese tiempo real tan buscado. Por lo que planteamos como objetivo para este TFG la aceleración de este tipo de algoritmos mediante diversos tipos de aceleradores usando OpenCL y de paso demostrar que OpenCL es una buena herramienta que permite códigos paralelizados con un gran Speedup a la par que funcionar en toda una diversa gama de dispositivos tan distintos como un GPU y una FPGA. Para lo anteriormente mencionado trataremos de desarrollar un código para cada algoritmo y optimizarlo de forma no especifica a una plataforma para posteriormente ejecutarlo sobre las diversas plataformas y medir tiempos y error para cada algoritmo. Para el desarrollo de este proyecto partimos de la teoría de dos algoritmos ya existentes: Lucas&Kanade monoescala y el Horn&Schunck. Además, usaremos estímulos para estos algoritmos muy aceptados por la comunidad como pueden ser el RubberWhale o los Grove, los cuales nos ayudarán a establecer la corrección de estos algoritmos y analizar su precisión, dando así un estudio referencia para saber cual escoger.

Soporte de sistema operativo para ahorro de energía en plataformas móviles con procesadores multicore asimétricos

Los procesadores multicore asimétricos con repertorio común de instrucciones (AMPsAsymmetric Multicore Processors) han sido propuestos recientemente como alternativa de bajo consumo a los procesadores multicore simétricos convencionales. Los AMPs combinan, en un mismo chip, cores rápidos de alto rendimiento, con cores más lentos y sencillos de consumo reducido. Uno de los ejemplos más destacados de procesador multicore asimétrico es el procesador big.LITTLE de ARM, que incorporan algunos modelos de teléfonos móviles y tablets disponibles en la actualidad. Trabajos previos han demostrado que para explotar los beneficios potenciales de los procesadores multicore asimétricos, el sistema operativo debe tener en cuenta el beneficio relativo (speedup) que cada aplicación experimenta al ejecutar en un core rápido frente a un core lento. Actualmente, los planificadores por defecto de los sistemas operativos de propósito general no tienen en cuenta la diversidad de speedups entre aplicaciones que puede estar presente en una carga de trabajo multiprogramada. En consecuencia, la asignación de aplicaciones a cores que hacen estos planificadores no extrae el máximo rendimiento por vatio de la plataforma. Recientemente se han realizado extensiones en el kernel Linux para ofrecer un mejor soporte de planificación en multicore asimétricos. Sin embargo, estas extensiones del planificador, utilizadas fundamentalmente en dispositivos móviles con el sistema operativo Android, tampoco tienen en cuenta la diversidad de speedups en las aplicaciones de la carga de trabajo. Por lo tanto estas extensiones no constituyen una aproximación robusta desde el punto de vista de la eficiencia energética. En este proyecto se lleva a cabo la evaluación exhaustiva de distintos algoritmos de planificación para multicore asimétricos sobre una plataforma provista de un procesador ARM big.LITTLE. El principal objetivo del estudio es cuantificar el grado de eficiencia energética y el rendimiento global proporcionado por implementaciones de estos algoritmos en el kernel Linux sobre hardware multicore asimétrico real.

Practical engineering of hard spin-glass instances

Recent technological developments in the field of experimental quantum annealing have made prototypical annealing optimizers with hundreds of qubits commercially available. The experimental demonstration of a quantum speedup for optimization problems has since then become a coveted, albeit elusive goal. Recent studies have shown that the so far inconclusive results, regarding a quantum enhancement, may have been partly due to the benchmark problems used being unsuitable. In particular, these problems had inherently too simple a structure, allowing for both traditional resources and quantum annealers to solve them with no special efforts. The need therefore has arisen for the generation of harder benchmarks which would hopefully possess the discriminative power to separate classical scaling of performance with size from quantum. We introduce here a practical technique for the engineering of extremely hard spin-glass Ising-type problem instances that does not require "cherry picking" from large ensembles of randomly generated instances. We accomplish this by treating the generation of hard optimization problems itself as an optimization problem, for which we offer a heuristic algorithm that solves it. We demonstrate the genuine thermal hardness of our generated instances by examining them thermodynamically and analyzing their energy landscapes, as well as by testing the performance of various state-of-the-art algorithms on them. We argue that a proper characterization of the generated instances offers a practical, efficient way to properly benchmark experimental quantum annealers, as well as any other optimization algorithm.

Background Traffic Modeling for Large-Scale Network Simulation

Network simulation is an indispensable tool for studying Internet-scale networks due to the heterogeneous structure, immense size and changing properties. It is crucial for network simulators to generate representative traffic, which is necessary for effectively evaluating next-generation network protocols and applications. With network simulation, we can make a distinction between foreground traffic, which is generated by the target applications the researchers intend to study and therefore must be simulated with high fidelity, and background traffic, which represents the network traffic that is generated by other applications and does not require significant accuracy. The background traffic has a significant impact on the foreground traffic, since it competes with the foreground traffic for network resources and therefore can drastically affect the behavior of the applications that produce the foreground traffic. This dissertation aims to provide a solution to meaningfully generate background traffic in three aspects. First is realism. Realistic traffic characterization plays an important role in determining the correct outcome of the simulation studies. This work starts from enhancing an existing fluid background traffic model by removing its two unrealistic assumptions. The improved model can correctly reflect the network conditions in the reverse direction of the data traffic and can reproduce the traffic burstiness observed from measurements. Second is scalability. The trade-off between accuracy and scalability is a constant theme in background traffic modeling. This work presents a fast rate-based TCP (RTCP) traffic model, which originally used analytical models to represent TCP congestion control behavior. This model outperforms other existing traffic models in that it can correctly capture the overall TCP behavior and achieve a speedup of more than two orders of magnitude over the corresponding packet-oriented simulation. Third is network-wide traffic generation. Regardless of how detailed or scalable the models are, they mainly focus on how to generate traffic on one single link, which cannot be extended easily to studies of more complicated network scenarios. This work presents a cluster-based spatio-temporal background traffic generation model that considers spatial and temporal traffic characteristics as well as their correlations. The resulting model can be used effectively for the evaluation work in network studies.

Distributed parallelization of greedy Mobile Network Optimization algorithms

The Mobile Network Optimization (MNO) technologies have advanced at a tremendous pace in recent years. And the Dynamic Network Optimization (DNO) concept emerged years ago, aimed to continuously optimize the network in response to variations in network traffic and conditions. Yet, DNO development is still at its infancy, mainly hindered by a significant bottleneck of the lengthy optimization runtime. This paper identifies parallelism in greedy MNO algorithms and presents an advanced distributed parallel solution. The solution is designed, implemented and applied to real-life projects whose results yield a significant, highly scalable and nearly linear speedup up to 6.9 and 14.5 on distributed 8-core and 16-core systems respectively. Meanwhile, optimization outputs exhibit self-consistency and high precision compared to their sequential counterpart. This is a milestone in realizing the DNO. Further, the techniques may be applied to similar greedy optimization algorithm based applications.

Dynamic Workload Parallelization and System Calibration on Single-Chip Cloud Computer

Many-core systems are emerging from the need of more computational power and power efficiency. However there are many issues which still revolve around the many-core systems. These systems need specialized software before they can be fully utilized and the hardware itself may differ from the conventional computational systems. To gain efficiency from many-core system, programs need to be parallelized. In many-core systems the cores are small and less powerful than cores used in traditional computing, so running a conventional program is not an efficient option. Also in Network-on-Chip based processors the network might get congested and the cores might work at different speeds. In this thesis is, a dynamic load balancing method is proposed and tested on Intel 48-core Single-Chip Cloud Computer by parallelizing a fault simulator. The maximum speedup is difficult to obtain due to severe bottlenecks in the system. In order to exploit all the available parallelism of the Single-Chip Cloud Computer, a runtime approach capable of dynamically balancing the load during the fault simulation process is used. The proposed dynamic fault simulation approach on the Single-Chip Cloud Computer shows up to 45X speedup compared to a serial fault simulation approach. Many-core systems can draw enormous amounts of power, and if this power is not controlled properly, the system might get damaged. One way to manage power is to set power budget for the system. But if this power is drawn by just few cores of the many, these few cores get extremely hot and might get damaged. Due to increase in power density multiple thermal sensors are deployed on the chip area to provide realtime temperature feedback for thermal management techniques. Thermal sensor accuracy is extremely prone to intra-die process variation and aging phenomena. These factors lead to a situation where thermal sensor values drift from the nominal values. This necessitates efficient calibration techniques to be applied before the sensor values are used. In addition, in modern many-core systems cores have support for dynamic voltage and frequency scaling. Thermal sensors located on cores are sensitive to the core's current voltage level, meaning that dedicated calibration is needed for each voltage level. In this thesis a general-purpose software-based auto-calibration approach is also proposed for thermal sensors to calibrate thermal sensors on different range of voltages.

Scalable unstructured mesh decomposition

As the efficiency of parallel software increases it is becoming common to measure near linear speedup for many applications. For a problem size N on P processors then with software running at O(N=P ) the performance restrictions due to file i/o systems and mesh decomposition running at O(N) become increasingly apparent especially for large P . For distributed memory parallel systems an additional limit to scalability results from the finite memory size available for i/o scatter/gather operations. Simple strategies developed to address the scalability of scatter/gather operations for unstructured mesh based applications have been extended to provide scalable mesh decomposition through the development of a parallel graph partitioning code, JOSTLE [8]. The focus of this work is directed towards the development of generic strategies that can be incorporated into the Computer Aided Parallelisation Tools (CAPTools) project.

Preventing encrypted traffic analysis

Many existing encrypted Internet protocols leak information through packet sizes and timing. Though seemingly innocuous, prior work has shown that such leakage can be used to recover part or all of the plaintext being encrypted. The prevalence of encrypted protocols as the underpinning of such critical services as e-commerce, remote login, and anonymity networks and the increasing feasibility of attacks on these services represent a considerable risk to communications security. Existing mechanisms for preventing traffic analysis focus on re-routing and padding. These prevention techniques have considerable resource and overhead requirements. Furthermore, padding is easily detectable and, in some cases, can introduce its own vulnerabilities. To address these shortcomings, we propose embedding real traffic in synthetically generated encrypted cover traffic. Novel to our approach is our use of realistic network protocol behavior models to generate cover traffic. The observable traffic we generate also has the benefit of being indistinguishable from other real encrypted traffic further thwarting an adversary's ability to target attacks. In this dissertation, we introduce the design of a proxy system called TrafficMimic that implements realistic cover traffic tunneling and can be used alone or integrated with the Tor anonymity system. We describe the cover traffic generation process including the subtleties of implementing a secure traffic generator. We show that TrafficMimic cover traffic can fool a complex protocol classification attack with 91% of the accuracy of real traffic. TrafficMimic cover traffic is also not detected by a binary classification attack specifically designed to detect TrafficMimic. We evaluate the performance of tunneling with independent cover traffic models and find that they are comparable, and, in some cases, more efficient than generic constant-rate defenses. We then use simulation and analytic modeling to understand the performance of cover traffic tunneling more deeply. We find that we can take measurements from real or simulated traffic with no tunneling and use them to estimate parameters for an accurate analytic model of the performance impact of cover traffic tunneling. Once validated, we use this model to better understand how delay, bandwidth, tunnel slowdown, and stability affect cover traffic tunneling. Finally, we take the insights from our simulation study and develop several biasing techniques that we can use to match the cover traffic to the real traffic while simultaneously bounding external information leakage. We study these bias methods using simulation and evaluate their security using a Bayesian inference attack. We find that we can safely improve performance with biasing while preventing both traffic analysis and defense detection attacks. We then apply these biasing methods to the real TrafficMimic implementation and evaluate it on the Internet. We find that biasing can provide 3-5x improvement in bandwidth for bulk transfers and 2.5-9.5x speedup for Web browsing over tunneling without biasing.

Algoritmos construtivos para treino de redes neuronais b-spline

Dissertação de mest. em Engenharia de Sistemas e Computação - Área de Sistemas de Controlo, Faculdade de Ciências e Tecnologia, Univ.do Algarve, 2001