PhD Defense - Ryan McCune
|Start:||3/23/2016 at 10:00AM|
|End:||3/23/2016 at 1:00PM|
|Location:||100 Stinson Remick|
Faculty and Students are welcome to attend the presentation portion of the defense.
Light refreshments will be served.
March 23, 2016
100 Stinson Remick
Adviser: Dr. Gregory Madey
Co-Adviser: Dr. Tim Weninger
Dr. Christian Poellabauer Dr. Laurel Riek Dr. Dong Wang
"DECENTRALIZATION IN DISTRIBUTED SYSTEMS FOR GRAPH PROCESSING AND MULTI-AGENT SIMULATION MODELING"
Centralization pervades conventional system architectures, from society to computing, where high-level decision-making is focused at a central coordinator. In the technology domain, centralization has led to performance bottlenecks and scalability challenges, as witnessed in distributed systems including Big Data processing and Unmanned Aerial Vehicles (UAVs). Techniques and approaches to relax centralized control in favor of decentralized methods are investigated.
Two graph processing projects will be discussed. For large-scale graph processing, centralized computational methods contribute to 2 notable challenges: i) impractical processing for distributed memory, and ii) a high data-access to computation ratio. In response to the first challenge, software frameworks have been developed as platforms for executing distributed algorithms, which decentralize computational scope to increase scalability. Insights from a comprehensive survey published on such frameworks will be presented. Next, an instance of reducing random data access for efficient graph traversal will be presented as part of research conducted while at IBM Zurich. Together, the projects illustrate principles for overcoming data-intensive computing challenges.