Estratégias de identificação paramétrica aplicadas à modelagem dinâmica de um servidor web Apache

Abstract

This article presents an experimental study about parametric identification techniques applied to the modeling of an Apache webserver. In order to simulate load variations at the server, an experimental arrangement was developed, which is composed of two personal computers, one used to run the Apache server and the other to generate workload by requesting services to the Apache. Auto-regressive (AR) parametric models were estimated at different operating points and workload conditions. The mean values of the MaxClients input (a parameter which is used to set the maximum number of the server's active processes) were used to define the operating points, in order to obtain the Apache server CPU utilization (in %) as output. 600 samples were collected at each operating point every 5 seconds. To proceed with the system identification, half of the data set was used for parameter estimation while the other half was used for model validation, at each operating point. A study of the most adequate system order showed that a 7th order model could be satisfactorily used for MaxClients low values operating points. However, the results showed that higher order models are needed for MaxClients higher values, due to system inherent non-linearities.