Automated Problem Diagnosis in Real-World Distributed Systems

This talk will focus on automated problem diagnosis for various real-world distributed enterprise systems. In particular, I will focus  on recent work that we have done to diagnose a range of real-world performance problems in production VoIP telecom systems, in HPC-oriented parallel file systems such as PVFS, Lustre and GPFS, as well as in MapReduce/Hadoop systems. The over-arching common theme of all of our research is to be able to localize performance problems in production environments, where one often has to use whatever instrumentation is available and cannot rely on the presence of labeled faulty data. I will also touch briefly on our work on problem diagnosis for embedded automotive systems and our related efforts in problem-visualization tools for system administrators.

About Priya Narasimhan
Prof. Priya Narasimhan is an Associate Professor with the Electrical & Computer Engineering Department, and the academic PI for the Intel Science and Technology Center in Embedded Computing (ISTC-EC) located at Carnegie Mellon. From 2010-11, she served as the Director of Intel Labs Pittsburgh. She is also currently the CEO/Founder of YinzCam, Inc., a Carnegie Mellon spin-off company focused on mobile large-scale live streaming in the high-density wireless environments of NHL/NFL stadiums. 

Her research has earned her an Alfred Sloan Fellowship, the 2009 Carnegie Science Center's Emerging Female Scientist Award, an NSF CAREER Award, the 2001 UCSB Lancaster Outstanding Doctoral Dissertation Award and two IBM Faculty Partnership Awards. Her teaching has earned her the 2008 Eta Kappa Nu (Carnegie Mellon Sigma Chapter) Excellence in Teaching Award, and Carnegie Mellon's College of Engineering's 2009 Benjamin Richard Teare Teaching Award. 

Her research interests lie in the fields of dependable distributed systems, fault-tolerance, embedded systems, mobile systems and sports technology.