Automated Performance Problem Determination by Observing Service Demand

Information Technology (IT) systems is a crucial step in the development
of Autonomic Computing.
A new approach to detect and determine performance problems in a
distributed computing system is proposed.
The approach is based on a simple principle that the value of a particular
performance metric, service demand, does not vary in time regardless of
external environment changes, such as transaction mix ratio changes and
intensity changes.
Thus an abrupt change of the service demand value suggests an occurrence
of a performance problem.
Based on this idea, the authors developed a system which detects and
localizes the machine causing a problem automatically by
estimating and observing the service demand value using network traffic
data flowing through the target IT system.
The developed system is suitable for large scale distributed computing
systems, since the system can monitor performance problems
simply by connecting to the mirror port of the network switch that
the machines of the target IT systems are connected to.
Verification experiment results for our system in two environments, a test
environment and a commercial production environment, show the
effectiveness of the system in detection and determination of performance
problems.