Network tomography for the Internet.
The Internet is a complex system consisting of 1000s of networks and 100,000s of routers. It exhibits exponential growth. Advances have been made in characterizing the workload and the performance through passive and active measurements. Network traffic traces are routinely taken at a handful of sites. In addition, several organizations routinely/sporadically transmit probes into the network and trace their resulting performance (loss, delay). These traces have been used to develop better insight on what current applications look like and how networks work. However, much remains to be done. There is no systematic way to correlate and organize observations made at different parts of the network. Thus, it is difficult to determine the behavior of various network elements, to determine the performance observed by applications, to efficiently isolate faults, to develop traffic models for applications such as napster and distributed games. I will attempt pose the challenges facing those of us interested in developing network tomography for the Internet. Here by network tomography I refer to a set of mechanisms that can be used to correlate observations from different parts of a network in order to describe network behavior that cannot be directly observed. I will motivate some of the challenges using measurements made at CAIDA [1]. I will ilustrate what is needed with examples from the MINC (multicast inference of network characteristics) project underway at UMass, AT&T, and LBNL [2,3]]. I will then pose a number of open problems.
1. Cooperative Association for Internet Data Analysis
2. MINC: Multicast-based Inference of Network-internal Characteristics
3. A. Adams, T. Bu, R. Caceres, N. Duffield, T.Friedman, J. Horowitz, F. Lo Presti, S.B. Moon, V. Paxson, D. Towsley, "The Use of End-to-end Multicast Measurements for Characterizing Internal Network Behavior." to appear in IEEE Communications Magazine, May 2000.