A Packet-level Characterization of Network Traffic

In this paper we show results from a packet-level traffic characterization aiming at finding spatial and temporal invariances of TCP based applications, such as HTTP and SMTP. We developed a methodology and a software architecture to build packet-level statistical characterization of network traffic based on large traces captured from wide area networks. In order to show the efficacy of the proposed packet-level approach, we applied our methodology to the traffic generated by applications running over HTTP - a typology of traffic that has been extensively studied in literature even if, as far as we know, no accurate packet-level characterization has been proposed as yet with regard to it - and to SMTP traffic. We analyzed traffic from high-speed access links of two different networks and, contrary to common beliefs, the results show properties of spatial and temporal invariance. The study of SMTP has been proposed to demonstrate the generalization of a packet-level approach. Indeed, results prove the general applicability of the methodology and, at the same time, how - also at packet level - different traffic shows different characterizations. Characterization results can be used in platforms for traffic simulation (like ns or ssfnet) and traffic generation (like D-ITG or MGEM).