Fingerprinting the Datacenter: Automated Classification of

Download Report

Transcript Fingerprinting the Datacenter: Automated Classification of

Fingerprinting the
Datacenter
Offense
Mykell Miller, Gautam Bhawsar
Crisis Detection
 Why don’t you detect unidentified crises?
 How severe is each crisis?
 May affect recovery method
 May affect detection probability and method
 Useful for operators to know
 Fingerprints are averaged over epochs in
the crisis
 Everywhere else you use the median
Time to identification
 What are you doing for 10 minutes?
 Why have the time between epochs be
15 minutes?
 Will shorter lead to more accurate or quicker
identification?
Data Center
 What is the configuration?
 Fat tree, etc.
 What kind of servers?
 Type of servers affects type and frequency
of crises
 Using only Windows
 Study cannot be applied to other OS
Data Set
 Only 19 examples
 8 types of crises had only 1 instance each
 Training set of only 5 examples
 No statistics on the user application
being run
 Don’t know if it’s representative