Thesis (M. S.)--Architecture, Georgia Institute of Technology, 2008. / Committee Chair: Spuybroek, Lars; Committee Member: Al-Haddad, Tristan; Committee Member: Romm, Stuart.

Thesis (Ph. D.)--Virginia Polytechnic Institute and State University, 1993. / Vita. Abstract. Includes bibliographical references (leaves 139-143). Also available via the Internet.

Amazon Spot Instances provide inexpensive service for high-performance computing. With spot instances, it is possible to get at most 90% off as discount in costs by bidding spare Amazon Elastic Computer Cloud (Amazon EC2) instances. In exchange for low cost, spot instances bring the reduced reliability onto the computing environment, because this kind of instance could be revoked abruptly by the providers due to supply and demand, and higher-priority customers are first served. To achieve high performance on instances with compromised reliability, Spark is applied to run jobs. In this thesis, a wide set of spark experiments are conducted to study its performance on spot instances. Without stateful replicating, Spark suffers from cascad- ing rollback and is forced to regenerate these states for ad hoc practices repeatedly. Such downside leads to discussion on trade-off between compatible slow checkpointing and regenerating on rollback and inspires us to apply multiple fault tolerance schemes. And Spark is proven to finish a job only with proper revocation rate. To validate and evaluate our work, prototype and simulator are designed and implemented. And based on real history price records, we studied how various checkpoint write frequencies and bid level affect performance. In case study, experiments show that our presented techniques can lead to ~20% shorter completion time and ~25% lower costs than those cases without such techniques. And compared with running jobs on full-price instance, the absolute saving in costs can be ~70%.

Every week there are over a billion new posts to Twitter services and many of those messages contain feedback to companies about their services. One company that has recognized this unused source of information is Netflix. That is why Netflix initiated the development of a system that will let them respond to the millions of Twitter and Netflix users that are acting as sensors and reporting all types of user visible outages. This system will enhance the feedback loop between Netflix and its customers by increasing the amount of customer feedback that is being received by Netflix and reducing the time it takes for Netflix to receive the reports and respond to them. The goal of the SPOONS (Swift Perceptions of Online Negative Situations) system is to use Twitter posts to determine when Netflix users are reporting a problem with any of the Netflix services. This work covers a subset of the meth- ods implemented in the SPOONS system. The volume methods detect outages through time series analysis of the volume of a subset of the tweets that contain the word “netflix”. The sentiment methods first process the tweets and extract a sentiment rating which is then used to create a time series. Both time series are monitored for significant increases in volume or negative sentiment which indicates that there is currently an outage in a Netflix service. This work contributes: the implementation and evaluation of 8 outage detection methods; 256 sentiment estimation procedures and an evaluation of each; and evaluations and discussions of the real time applicability of the system. It also provides explanations for each aspect of the implementation, evaluations, and conclusions so future companies and researchers will be able to more quickly create detection systems that are applicable to their specific needs.

In order to overcome the difficulties encountered in closed form solutions, moment-curvature-thrust relationships are developed for hybrid and nonhybrid cross sections utilizing an open form method. The use of horizontal sectors permits the inclusion of residual stresses and/or nonbilinear stress-strain relationships, if desired. Theoretical and experimental data are compared. Applications to circular tubes and other cross sections are discussed. Results indicate that open form solutions are feasible for calculating moment-curvature-thrust data. Hybrid cross sections are easily treated by open form solutions.

Structural and functional brain images are generated as essential modalities for medical experts to learn about the different functions of the brain. These images are typically visually inspected by experts. Many software packages are available to process medical images, but they are complex and difficult to use. The software packages are also hardware intensive. As a consequence, this dissertation proposes a novel Neuroimaging Web Services Interface (NWSI) as a series of processing pipelines for a common platform to store, process, visualize and share data. The NWSI system is made up of password-protected interconnected servers accessible through a web interface. The web-interface driving the NWSI is based on Drupal, a popular open source content management system. Drupal provides a user-based platform, in which the core code for the security and design tools are updated and patched frequently. New features can be added via modules, while maintaining the core software secure and intact. The webserver architecture allows for the visualization of results and the downloading of tabulated data. Several forms are ix available to capture clinical data. The processing pipeline starts with a FreeSurfer (FS) reconstruction of T1-weighted MRI images. Subsequently, PET, DTI, and fMRI images can be uploaded. The Webserver captures uploaded images and performs essential functionalities, while processing occurs in supporting servers. The computational platform is responsive and scalable. The current pipeline for PET processing calculates all regional Standardized Uptake Value ratios (SUVRs). The FS and SUVR calculations have been validated using Alzheimer's Disease Neuroimaging Initiative (ADNI) results posted at Laboratory of Neuro Imaging (LONI). The NWSI system provides access to a calibration process through the centiloid scale, consolidating Florbetapir and Florbetaben tracers in amyloid PET images. The interface also offers onsite access to machine learning algorithms, and introduces new heat maps that augment expert visual rating of PET images. NWSI has been piloted using data and expertise from Mount Sinai Medical Center, the 1Florida Alzheimer’s Disease Research Center (ADRC), Baptist Health South Florida, Nicklaus Children's Hospital, and the University of Miami. All results were obtained using our processing servers in order to maintain data validity, consistency, and minimal processing bias.

<p>As the number of transistors that are integrated onto a silicon die continues to in- crease, the compute power is becoming a commodity. This has enabled a whole host of new applications that rely on high-throughput computations. Recently, the need for faster and cost-effective applications in form-factor constrained environments has driven an interest in on-chip acceleration of algorithms based on Monte Carlo simula- tions. Though Field Programmable Gate Arrays (FPGAs), with hundreds of on-chip arithmetic units, show significant promise for accelerating these embarrassingly paral- lel simulations, a challenge exists in sharing access to simulation data amongst many concurrent experiments. This thesis presents a compute architecture for accelerating Monte Carlo simulations based on the Network-on-Chip (NoC) paradigm for on-chip communication. We demonstrate through the complete implementation of a Monte Carlo-based image reconstruction algorithm for Single-Photon Emission Computed Tomography (SPECT) imaging that this complex problem can be accelerated by two orders of magnitude on even a modestly-sized FPGA over a 2GHz Intel Core 2 Duo Processor. Futhermore, we have created a framework for further increasing paral- lelism by scaling our architecture across multiple compute devices and by extending our original design to a multi-FPGA system nearly linear increase in acceleration with logic resources was achieved.</p> / Master of Applied Science (MASc)

The interconnect mechanisms (shared bus or crossbar) used in current chip-multiprocessors (CMPs) are expected to become a bottleneck that prevents these architectures from scaling to a larger number of cores. Tiled CMPs offer better scalability by integrating relatively simple cores with a lightweight point-to-point interconnect. However, such interconnects make snooping impractical and, thus, require alternative solutions to cache coherence. This thesis proposes a novel, cost-effective hardware mechanism to support shared-memory parallel applications that forgoes hardware maintained cache coherence. The proposed mech- anism is based on the key ideas that mapping of lines to physical caches is done at the page level with OS support and that hardware supports remote cache accesses. It allows only some controlled migration and replication of data and provides a sufficient degree of flexibility in the mapping through an extra level of indirection between virtual pages and physical tiles. The proposed tiled CMP architecture is evaluated on the SPLASH-2 scientific benchmarks and ALPBench multimedia benchmarks against one with private caches and a distributed direc- tory cache coherence mechanism. Experimental results show that the performance degradation is as little as 0%, and 16% on average, compared to the cache coherent architecture across all benchmarks for 16 and 32 processors.

As a throughput-oriented device, Graphics Processing Unit(GPU) has already integrated with cache, which is similar to CPU cores. However, the applications in GPGPU computing exhibit distinct memory access patterns. Normally, the cache, in GPU cores, suffers from threads contention and resources over-utilization, whereas few detailed works excavate the root of this phenomenon. In this work, we adequately analyze the memory accesses from twenty benchmarks based on reuse distance theory and quantify their patterns. Additionally, we discuss the optimization suggestions, and implement a Bypassing Aware(BA) Cache which could intellectually bypass the thrashing-prone candidates. BA cache is a cost efficient cache design with two extra bits in each line, they are flags to make the bypassing decision and find the victim cache line. Experimental results show that BA cache can improve the system performance around 20\% and reduce the cache miss rate around 11\% compared with traditional design.

Mobile devices such as tablets and smartphones are becoming more common, and they are holding more information. This includes private information such as contacts, financial data, and passwords. At the same time these devices have network capability with access to the Internet being a prime feature. Little research has been done in observing the network traffic produced by these mobile devices. To determine if private information was being transmitted without user knowledge, the mobile capture lab and a set of procedures have been created to observe, capture and analyze the network traffic produced by mobile devices. The effectiveness of the lab and procedures has been evaluated with the analysis of four common mobile devices. The data analyzed from the case studies indicates that, contrary to popular opinion, very little private information is transmitted in clear text by mobile devices without the user’s knowledge.