A technique for distributed systems performance evaluation and control
Sunil Santha, Udo W. Pooch · 1997
A methodology is presented, with which the performance of a distributed system can be established within an empirical framework. In this methodology (JAGATH), a set of performance variables representing the system performance is derived in terms of a combination of a set of performance measures. These performance measures represent the actual measurements of the events in the system. The system is viewed in terms of several levels of abstractions. Of the event changes that can be described at each of these levels, the events closest in representation to the performance objective are selected for measurement. The set of large amount of performance measures hence obtained is analyzed using principal component analysis to produce a smaller number of independent and orthogonal performance variables, exposing the structure and information in the original measurements. These performance variables are used to display the system performance status in the form of Kiviat graphs. The causal relationship between the external performance variables and the internal system control variables and the workload characteristics can be established. This can be used in a performance 'tuning' system where performance deteriorations of the system indicated by undesirable Kiviat patterns are counteracted by corresponding changes in the internal system control and the workload. A prototype implementation of a tool utilizing the aforementioned technique is presented and the results discussed.