Bell Labs Internet Traffic Research

Measurement Analysis Modeling Generation



* Goals:
We measure, study, and model the characteristics of Internet traffic for device design, network performance monitoring, and traffic engineering.
* Three Fundamental Areas of Work
Our tool development emanates from three fundamental areas of work: (1) packet header data collection projects (2) hardware and software environment for analyzing very large packet header databases (3) fundamental research in the characterization and modeling of traffic.


Statistical Multiplexing: Mathematics Takes Over the Internet

True, network devices put packets on Internet links and multiplex the packets from different simultaneous active connections, but then the mathematical and statistical laws of stochastic processes take over. Extensive empirical and theoretical studies of detailed packet data, including inter-arrivals and sizes, reverse the commonly-held belief that Internet traffic is everywhere bursty. They also demonstrate that the magnitude of statistical multiplexing has a dramatic effect on traffic characteristics and needs to become part of the fundamental conceptual framework that guides the study of Internet traffic.

Left to Right: Dong Lin, Don Sun (standing), Bill Cleveland, and Jin Cao.

The technical details are given in four papers by the group in the photo. ``Internet Traffic Tends Toward Poisson and Independent as the Load Increases'', Nonlinear Estimation and Classification , eds. C. Holmes, D. Denison, M. Hansen, B. Yu, and B. Mallick, Springer, New York, 2002. ``Internet Traffic: Statistical Multiplexing Gains'', DIMACS Workshop on Internet and WWW Measurement, Mapping and Modeling, 2002. ``On the Nonstationarity of Internet Traffic'', Proc. ACM SIGMETRICS `01, 102-112, 2001. ``The Effect of Statistical Multiplexing on the Long Range Dependence of Internet Packet Traffic'', Bell Labs Tech Report (Very Detailed), 2002 . See Papers Available on the Web .

Lucent Technologies, committed to Internet measurement and traffic engineering to increase the efficiency of Internet devices and core networks, put out a Press Release of the traffic discoveries on the business wires.

INFORMATIONWEEK.COM wrote a News Story, including a trading of opinions with Steve Coya, Executive Director of the Internet Engineering Task Force.


* Overviews
Packet Header Measurement
TCP/UDP/IP packet header collection, and organization of the headers into connection flows, is a powerful measurement framework for Internet traffic study. Our two packet collection projects have yielded nearly 1 terabyte of compressed headers that have been moved into our S-Net environment for traffic analysis. In addition, our database for analysis includes headers from other collection projects such as NLANR and NZIX.
S-Net
This open-source UNIX software system, based on the S language for graphics and data analysis, is allowing analyses of very large Internet packet-header databases both in depth and breadth.
Fundamentals of Traffic Modeling
Our approach exploits the fundamental structure of Internet traffic, the superposition of different traffic sources. Superposition underlies our study of the data and underlies our statistical models. It has led to a number of important discoveries about the fundamental nature of Internet traffic, and to realistic, parsimonious models for traffic variables.
PackMime: Statistical Modeling of Connection Request Variables
In this large-scale project, which requires attention to each Internet application studied, we are carrying out a comprehensive statistical modeling of connection request variables. The models are needed both for a fundamental understanding of Internet traffic, and for providing traffic request generation for network simulators such as NS.
Helios Next-Generation-Internet Project
As part of this DARPA funded work our Helios team from Bell Labs Research, Lucent Optical Network Group, NCNI Research, and UNC Computer Science, is collecting and analyzing high-speed Internet traffic data on the NCNI gigapop, and we are implementing a S-Net-Helios environment at link speeds of up to 2.5 gigabits/second.
Visualization
Our studies of traffic depend heavily on data visualization We use a number of existing tools and have developed new ones.


* Papers Available On the Web