Failure Prediction with Exact Localization

• funded by: LBNL
• funding level: $84,684
• duration: 10/18/2016 - 08/15/2017 (no-cost extension to 3/31/2018)

The objective of this work is to assess the potential of machine learning techniques for pin-pointing failures before they happen with high true positive and low false positive rates.

Publications:

• "Doomsday: Predicting Which Node Will Fail When on Supercomputers" by Anwesha Das, Frank Mueller, Paul Hargrove, Eric Roman, Scott Baden, in Supercomputing (SC), Nov 2018, pages (accepted), Best Paper Candidate.
• Desh: Deep Learning for System Health Prediction of Lead Times to Failure in HPC by Anwesha Das, Frank Mueller, Charles Siegel, Abhinav Vishnu in High-Performance Parallel and Distributed Computing (HPDC), Jun 2018, pages 40-51.

Posters:

• Holistic Root Cause Analysis of Node Failures in Production HPC. Anwesha Das, advised by Frank Mueller. Referred ACM SRC Poster at Supercomputing (SC'18) Nov 2018.
• Aarohi: Automaton-based Low-cost Online Failure Prediction . Anwesha Das, Frank Mueller. Referred ACM SRC Poster at ASPLOS'18, Mar 2018.
• Desh: Deep Learning for HPC System Health Resilience. Anwesha Das, Abhinav Vishnu, Charles Siegel, Frank Mueller. Referred Poster at Supercomputing (SC'17) Nov 2017.
• Pin-pointing Node Failures in HPC Systems. Anwesha Das, Frank Mueller, Paul Hargrove, Eric Roman. Referred Poster at Supercomputing (SC'16) Nov 2016.

Theses:

• "Predicting Location and Time of Anomalies in Large-Scale Computing Systems via LogMining" by A. Das, Ph.D. Thesis, North Carolina State University, Aug 2019