Facilitating Active Network Monitoring of NWSC’s Yellowstone

08/02/2012 - 6:00pm
FL-1001, Small Seminar Room
Brendan Sheridan


We explore the use of active network monitoring on hardware that is similar to the Yellowstone system that is currently being installed in NWSC. Active monitoring of a supercomputer can potentially identify specific communication patterns employed by users as well as the placement of jobs by the system’s scheduler and the resulting shared links. This presentation seeks to cover the available tools and associated difficulties in active monitoring of a Mellanox Infiniband based network.  We illustrate the measured network statistics generated from several common yet problematic communication patterns.