Methodologies and analyses of broadband access network traffic
Amit Kumar Sinha · 2007
The introduction of technologies such as Digital Subscriber Line (DSL), Broadband Fixed Wireless (BFW), and Cable modems in mile access networks has enabled the development of user applications that require large bandwidth along with low delay and jitter. Some of these applications include streaming audio/video, online games, Internet Protocol (IP) telephony and peer-to-peer (P2P) file sharing. While a variety of applications have surfaced in the last decade that challenge the new capabilities provided by broadband access network, not much has been done to understand the nature of such traffic in view of their impact on the access. The common belief is that users on one broadband access generate traffic that is not intrinsically different from users on other broadband accesses with different underlying technologies. This belief, if incorrect, could lead to suboptimal or bad network design. Moreover, there has been no analysis using real traffic data to verify this assumption; one reason being that such data is hard to get. In this dissertation we introduce post-processing methodologies to generate meaningful information to be analyzed, from the scattered raw data sets provided to us by Internet Service Providers (ISPs) that were collected at different times using different data collection tools. The first part of our analysis shows that the Transmission Control Protocol (TCP) connection arrival process depends on the underlying Medium Access Control (MAC) protocol and the error control mechanisms, which differ for different access schemes. Since the network resource allocation problem arises at the connection level, the design of a broadband access network should take into account the influence of the underlying technology on the user traffic. The second part of our analysis is focussed on understanding the behavior of an application (a network game) with stringent QoS (Quality of Service) requirements on broadband access network. We found that the latency (QoS requirement) was well within the boundaries suggested for this application. Our analysis shows that the user duration is no longer exponential but rather heavy tailed, which reflects a change in user behavior with broadband access that accounts for the popularity of such applications.