In Search Of Determinants Of HTTP Traffic Self-Similarity On A Web Server

The self-similar nature of Web traffic has been widely recognized, but factors influencing its degree remain an open research question. Our paper investigates relationships between Web traffic self-similarity and various features of HTTP traffic at the input of an e-commerce Web server. Using real server log data, self-similarity was quantified through the Hurst parameter, which was estimated with the use of various methods. Web traffic burstiness was analyzed, and correlations between the average Hurst parameter and selected Web traffic features were examined. Results confirm that the analyzed Web traffic exhibits self-similarity across different time scales. The degree of burstiness and traffic intensity varies significantly throughout the day. In particular, the nighttime traffic is less intensive and not bursty whereas the traffic observed during afternoon and evening hours is characterized by increased user activity and burstiness. Results show a weak negative correlation between self-similarity and both burstiness parameters and request arrival rates, suggesting that high traffic intensity and burstiness do not necessarily correspond to increased self-similarity. In contrast, there is a weak positive correlation between self-similarity and the proportion of Web bot requests, particularly for marketing bots, while search engine bots exhibited no significant relationship with self-similarity. The findings suggest that some bot activity may contribute to Web traffic self-similarity.