Abstract

Debugging Bad Performance in Huge Infrastructure: Using ML and AI (Presentation) Page 1 Debugging Bad Performance in Huge Infrastructure: Using ML and AI Mariam Kiran (ESnet) Anirban Mandal (RENCI) George Papadimitriu (USC) Cong Wang (RENCI) Ewa Deelman (USC) Huge Data Workshop 13-14 April 2020 1 Page 2 Complex Workflows using Distributed Facilities • Distributed workflows massive computations and large data movements • Workflow performance: – debug performance – performance optimization • Infrastructure Troubleshooting: – Anomaly finding – root cause analysis 2 Workflow Performance analysis Infrastructure Troubleshooting Page 3 Challenge: Working with Distributed Facilities • Current infrastructures are not designed to diagnose bad performance in distributed settings • Distributed performance data repositories • RQ: Can we use ML/AI to build “normal” profiles to debug …

Date

2020

Authors

Mariam Kiran, Anirban Mandal, George Papadimitriu, Cong Wang, Ewa Deelman