Machine Learning and Knowledge Discovery for Engineering Systems Health Management

CHAPTER 10

Real-Time Identification of Performance Problems in Large Distributed Systems

Moises GoldszmidtDawn WoodardPeter Bodik

CONTENTS

10.1  Introduction

10.2  Problem Definition

10.3  From Collected Signals to Fingerprints

10.3.1  Summarizing the State of the Datacenter

10.3.2  Crisis Modeling

10.3.3  Selecting the Relevant Signals

10.4  Identifying a Crisis

10.4.1  Cluster Modeling

10.4.2  Computing the Probability of the Crisis Label

10.4.3  Prior Specification

10.5  Experiments and Results

10.5.1  System Under Study and Data

10.5.2  Experiments

10.5.3  Offline Clustering

10.5.4  Operational Setting

10.6  Discussion

10.7  Conclusions

Acknowledgments

References

10.1  INTRODUCTION

The large networked distributed systems that support ...