KIT | KIT-Bibliothek | Impressum | Datenschutz

Digital Artefacts and Appendix for the Dissertation of Moritz v. Looz

Looz, Moritz von

Abstract (englisch):
The corresponding dissertation contains several types of experiments. The graph generation experiments were conducted on a shared-memory machine with 16 cores. To reproduce these experiments, this package contains a docker image. For the experiments regarding large-scale graph partitioning, we used phase 1 of the SuperMUC cluster, including tens of thousands of cores. For this set of experiments, we do not include a script to reproduce it, as the original computing environment is no longer available and the experimental pipeline depends on details of the system setup, making an automated reproduction difficult. Instead, we offer the output logs and summarized data of our experiments.

Open Access Logo


Zugehörige Institution(en) am KIT Institut für Theoretische Informatik (ITI)
Publikationstyp Forschungsdaten/Bilder
Jahr 2019
Erstellungszeitraum 18.06.2019
Identifikator DOI (KIT): 10.5445/IR/1000095237
KITopen-ID: 1000095237
Lizenz CC BY-NC-ND 4.0: Creative Commons Namensnennung – Nicht kommerziell – Keine Bearbeitungen 4.0 International
Projektinformation SPP 1736 (DFG, DFG KOORD, ME 3619/3-2)
Wave-Sim - TP: ITI (BMBF, 01IH15004B)
Schlagworte graph generation, hyperbolic geometry, random hyperbolic graphs, k-means, graph partitioning, probabilistic query, distributed computing, load balancing
Liesmich

There are two ways to reproduce the graph generation experiments:

  1. Unpack the archive hyperbolic-scripts.zip, run the python script download-install.py, followed by the script experiment-plot.py. Install any missing dependencies, rerun if necessary.
  2. Use "docker load hyperbolic-image.tar" to load the image and "docker run hyperbolic-reproducibility" to run it. After the experiments are completed, use docker save -o filesystem.tar. to save the filesystem of the docker image to a tarball, inspect the compiled plots in /app. Running all the experiments might take ~100 hours, 256 GiB of memory and 16 cores.

The same methods apply to reproduce the experiments to partition protein graphs, with the script archive protein-scripts.zip and the docker image protein-image.tar. These experiments take about 3 hours and 4 GiB of memory.

The experimental logs of the partitioning with balanced k-means are found in the archive Geographer-comparison-log-files.tar.gz.

Art der Forschungsdaten Dataset
Relationen in KITopen
KIT – Die Forschungsuniversität in der Helmholtz-Gemeinschaft
KITopen Landing Page