Spatial Econometrics Models in Web Server’s Performance

In recent years we saw, how desirable is possibility of mobile communications in modern society. The consequence of this is to have Internet more reliable and predictable in context of Web access. Thus, there is a need to analyzing Web server’s performance and trying to predict future demand on given server’s. This kind of research requires spatial methods of analysis of such data. Therefore we decided using spatial econometrics methods to explore Web server’s performance.

This paper contains description the spatial regression models: Classic Regression Model (CRM), Spatial Lag Model (SLM) and Spatial Error Model (SEM). We use these models to predict total download time of data from Web servers. The real-life dataset was obtained in active experiments performed by the Multiagent Internet Measurement System (MWING), which monitored web transactions issued by MWING’s agent located in Gdańsk, Poland and targeting Web servers in Europe. Data analyzed in this paper contains the measurements, which were taken every day at the same time: at 6:00 a.m., 12:00 a.m. and 6:00 p.m. We presented our analysis of measurement data and created spatial econometric models. Next, influences on prediction errors in regression models were described. After that we compared econometric with geostatistical methods. At the end, conclusions and future research directions to Web performance predictions were given.