SIParCS 2018- Todd Yoder

Supercomputer Infiniband Fabric Analysis

Recorded Talk 

Supercomputing centers strive to provide productive and efficient services to users. Static analysis and performance monitoring are useful tools which can help supercomputer teams maintain and improve existing network fabrics. This project focuses on developing graph theory analysis tools in C++ for supercomputers which use the InfiniBand architecture. InfiniBand is a computer-networking communications standard used commonly in high-performance computing and is currently the architecture used by each supercomputer operated by the National Center for Atmospheric Research. The analysis tools developed in this project are designed as plugins for Tulip, an open source data visualization program. Tulip provides powerful visualization of the supercomputer network and a number of useful functions utilized by the plugins. The plugins developed in this project include “Random Nodes”, “Regularity Test”, “Node on Cycle Test”, “Geodesic Test”, “Bipartite Test”, and “Degrees Min Max”. “Random Nodes” selects two random nodes in the graph. “Regularity Test” determines if every node has the same degree. “Node on Cycle Test” tests if the selected node lies on a closed loop. “Geodesic Test” checks if a selected path is the shortest between the two end nodes.  “Bipartite Test” determines if the nodes of the graph can be partitioned into two subsets such that every edge in the graph joins the two subsets. Finally, “Degrees Min Max” returns the smallest and largest node degrees. These plugins provide basic graph theory tools, building a foundation for further InfiniBand analysis in Tulip and equipping supercomputer centers with additional means for improving services provided to users.

Mentors: Nathan Rini, Tom Kleespies

Slides