Computer Science and Engineering, Department of


First Advisor

Hongfeng Yu

Second Advisor

Lisong Xu

Third Advisor

Chi Zhang

Date of this Version

Summer 8-1-2019


Yves Tuyishime(2019). Distributed Edge Bundling for Large Graphs (Master's Thesis). University of Nebraska-Lincoln.


A THESIS presented to the faculty of the Graduate College at the University of Nebraska In partial fulfillment of requirements for the degree of Master of Science (Computer Science), under the supervision of Professor Hongfeng Yu. Lincoln, Nebraska: August, 2019

Copyright(c) 2019 - Yves Tuyishime


Graphs or networks are widely used to depict the relationships between data entities in diverse scientific and engineering applications. A direct visualization (such as node-link diagram) of a graph with a large number of nodes and edges often incurs visual clutter. To address this issue, researchers have developed edge bundling algorithms that visually merge similar edges into curved bundles and can effectively reveal high-level edge patterns with reduced visual clutter. Although the existing edge bundling algorithms achieve appealing results, they are mostly designed for a single machine, and thereby the size of a graph they can handle is limited by the available memory of the machine.

To tackle large-scale graphs, a more scalable solution is to carry out edge bundling using multiple machines in a distributed environment. However, the existing edge bundling algorithms typically require the global information structure of a graph. Therefore, with a simple division of the edges of a graph, it is challenging to balance the workload and lower inter-processor communication among the processors. In this work, we select a representative edge bundling algorithm, Force-Directed Edge Bundling (FDEB), and parallelize it in a distributed environment. Particularly, to address the difficulties of partitioning and distributions of a large graph among processors, we first create a high dimensional space to represent the data distribution of a graph in FDEB. Second, we map each edge as a data point in this high dimensional space, and then partition and distribute the point cloud among processors. In this way, we can significantly reduce the data communication across processors, and ensure each processor assigned with a similar workload. We demonstrate the scalability of our distributed algorithm in our experimental study. Although we design the distributed approach for FDEB, we expect that the parallelization methodology developed in this work can be extended to other edge bundling algorithms as well.

Adviser: Hongfeng Yu