Visualization of Network Traffic and Acceleration of Linear Solutions Using Graphics Processors
Lee PeriolatOral Defence Date:
Monday, July 30, 2018 - 15:00Location:
Prof. Bill Hsu
This report documents three components that comprise the overall Culminating Experience Project. The first component is City to City, an interactive exhibit that inspires viewers to think about their internet traffic from new perspectives. City to City’s user traffic is traced and geolocated: the physical locations of the data routing devices are found from their internet protocol (IP) addresses, and the data path is animated on a world map, with sound accompaniment, in real-time. City to City was on view at the Zero One Garage in San Jose, in Spring 2014. The second component is a detailed performance study of Artisynth, a 3D biomechanical simulation platform that can be used to create compute-intensive models of anatomical structures. We studied the use of graphics processing units (GPU) to accelerate Artisynth computations. We ported Artisynth components to the GPU, using the Cusp and Cusolver libraries, and performed detailed measurements and characterizations of the code behavior. The third component is a detailed performance study of Diffusion MRI (dMRI) codes in tractography computations. Fiber tracking algorithms can be applied to dMRI data to generate connectomes, or maps of neural connections in the brain. We ran preliminary experiments that located performance bottlenecks in tractography calculations. We found the main bottleneck was to be the solution of a large set of linear equations; tractography codes may also benefit from GPU-based acceleration.