No abstract

Recently, measurement based studies of software systems prolifirated, reflecting an increasingly empirical ficus on system availability, reliability, aging and fault tolerance. However, it is a non-trivial, error-prone, arduous, and time-consuming task even for experienced system administrators and statistical analysts to know what a reasonable set of steps should include to model and success-,fully predict performance variables or system failures of a complex software system. Reported results are fragmented and focus on applying statistical regression techniques to captured numerical system data. In thir pap~r, we propose a best practice guide for building empirical models based on our experience with forecasting Apache web sewer performance variables and forecasting call availability of a real world telecommunication system. To substantiate the presented guide and to demonstrate our approach step-by-step we model and predict the response time and the amount of free physical memory of an Apache web sewer system. Additionally, we present concrete results for a) variable selection where we cross benchmark three procedures, b) empirical model building where we cross benchmark four techniques and c) sensitivity analysis. This besr practice guide intends to assist in configuring modeling approaches systematically .for best estimation andprediction results.

No abstract

The quality of Radial Basis Functions (RBF) and other nonlinear learning networks such as Multi Layer Perceptrons (MLP) depend significantly on issues in architecture, learning algorithms, initialisation heuristics and regularization techniques. Little attention has been given to the effect of mixture transfer functions in RBF networks on model quality and efficiency of parameter optimisation. We propose Universal Basis Functions (UBF) with flexible activation functions which are parameterised to change their shape smoothly from one functional form to another. This way they can cover bounded and unbounded subspaces depending on data distribution. We define UBF and apply them to a number of classification and function approximation tasks. We find that the UBF approach outperforms traditional RBF with the Hermite data set, a noisy Fourier series and a non φ-separable classification problem, however it does not improve statistically significant on the Mackey-Glass chaotic time series. The paper concludes with comments and issues for future research.