University of North Texas
Towards Dynamic Network Analysis: Benchmarks, Applications and Algorithms
Abstract
dc:descriptionGraphs are a fundamental abstraction for modeling complex systems, representing entities as vertices and relationships as edges across domains ranging from social networks and biological systems to High-Performance Computing infrastructures. Most real-world graphs are dynamic in nature, continuously evolving over time. This evolution alters critical graph properties such as shortest paths, and strongly connected components, demanding computational methods that can efficiently adapt to these changes at scale. At the same time, the growing scale and diversity of dynamic networks introduces additional demands: the research community needs realistic benchmarks to evaluate dynamic graph algorithms under conditions that reflect real-world behavior, and new application domains continue to emerge where dynamic graph analysis can provide actionable insights that existing tools cannot. Advancing the study of dynamic networks therefore requires progress on three inter- connected fronts: realistic benchmarks that capture temporal evolution, novel applications that demonstrate the broader utility of dynamic graph analysis, and scalable algorithms that maintain graph properties efficiently as networks evolve. This dissertation presents three contributions towards dynamic network analysis. First, we introduce GraTiS-Gen, a synthetic streaming network generator that models the temporal evolution of real-world networks by capturing structural properties such as triadic closure and network size, providing realistic and reproducible benchmarks for evaluating dynamic graph algorithms. Second, we present GraphIOX, a graph-based framework that models process-file interactions as evolving graphs to identify I/O bottlenecks and guide optimization in High-Performance Computing workloads. Third, we propose a parallel in- cremental algorithm for maintaining Strongly Connected Components in evolving directed graphs, introducing a compact metagraph representation and hub identification technique that together resolve up to 97% of edge insertions without explicit graph traversal, achieving up to 5.5×speedup over state-of-the-art static SCC algorithms on real-world graphs.
Degree
thesis:*- Grantor dc:publisher
- University of North Texas
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Pandey, Aashish
- Contributors dc:contributor
-
- Bhowmick, Sanjukta
- Ji, Yuede
- Wendt, Jeremy D.
- He, Yanyan
- Fu, Song
Subjects
dc:subject × 3Rights
dc:rights- Statement dc:rights
-
- Public
- Pandey, Aashish
- Copyright
- Copyright is held by the author, unless otherwise noted. All rights Reserved.
- Language dc:language
- English
Identifiers
dc:identifier.*- Identifier
-
local-cont-no: submission_5090
ark: ark:/67531/metadc2652575