Automatic Fault Diagnosis in Cloud Infrastructure
Authors:
- Qian ZhuQZQian Zhu
- Teresa TungTTTeresa Tung
- Qing XieQXQing Xie
- 1 December 2013
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- Vol. 1, 467-474
Abstract
With cloud computing, a cycle of fault diagnosis and recovery becomes the norm. There is a large amount of monitoring data and log events available, but it is hard to figure out which events or metrics are critical in fault diagnosis. Other approaches model faults as a deviation from normal behaviors, and thus are less applicable in cloud where changes in the environment may impact what is considered normal. In this work, we propose an adaptive and flexible fault diagnosis framework to automatically identify the key fault indicators and detect fault patterns. Leveraging ideas from social media, we represent the hierarchical relationships among metrics and events as well as how they relate to faults. We apply the EdgeRank algorithm to decide the key events that contribute to a fault. Our approach works across different environments to detect the potential faults. We evaluated our framework using a cloud-based enterprise system using a list of injected faults that vary from environmental (e.g. virtual machine or network) to application degradation. We considered both private and public clouds. Our solution achieves over 90% detection accuracy with modest overhead. A comparison of our approach shows it is more accurate than alternative approaches in the literature.Keywords
This publication has 13 references indexed in Scilit:
- Automatic fault characterization via abnormality-enhanced classification2012
- Draco: Statistical diagnosis of chronic problems in large distributed systems2012
- Detecting Abnormal Machine Characteristics in Cloud Infrastructures2011
- Efficient Fault Detection and Diagnosis in Complex Software Systems with Information-Theoretic MonitoringIEEE Transactions on Dependable and Secure Computing, 2011
- Extracting Overlay Invariants of Distributed Systems for Autonomic System Management2010
- Fingerprinting the datacenter2010
- Guided Problem Diagnosis through Active Learning2008
- A comparative study of pairwise regression techniques for problem determinationProceedings of the 2007 conference of the center for advanced studies on Collaborative research - CASCON '07, 2007
- SIGNAL DETECTION THEORY: VALUABLE TOOLS FOR EVALUATING INDUCTIVE LEARNING1989
- A Mathematical Theory of CommunicationBell System Technical Journal, 1948