Tracking a Metamorphic Infrastructure: observations on our (in)ability to accurately predict, analyze or even measure conditions on the global Internet
K.C. Claffy, Cooperative Association for Internet
Data Analysis (CAIDA), USA
We describe the state-of-the-art in the analysis and visualization of Internet data, using four primary categories: topology, workload, performance and routing. For each area of measurement, we assess current conditions limiting our ability to analyze, correlate, or visualize core Internet infrastructure data. We also highlight priorities for expanding insights into traffic and routing behaviour and enhancing our ability to measure and analyze the effects of new technology, protocols, and applications on the global Internet.
Topology data can describe network link infrastructure at a variety of layers. Topology measurement and analysis can reveal macroscopic characteristics of the global Internet, e.g., 'how big and what constitutes the Internet 'core'?', as well as provide parameters to topology-generator models.
Workload measurements involve the passive monitoring of traffic as it traverses a link or router, and allows analysis of distributions of traffic protocol, packet size, interarrival times, geographic flow, and per-user bandwidth consumption.
Performance measurements typically involve active probing of traffic into the network to assess latency or throughput characteristics of paths. Broad scale latency measurements (i.e., to thousands of relevant hosts) at reasonable granularity can provide a database for isolating global problems within the infrastructure, as well as assessing service quality by country or other granularity of interest.
Routing data includes analysis from border gateway protocol (bgp) routing tables, which reflect relationships between individual autonomous systems (ases) at a given point in time.
We also show instances of correlation among different measurement types, e.g., how actual (probed) topology differs from bgp-articulated topology (not to mention shortest path); how latency and throughput relate to other measures of path 'length'; how workload and routing changes affect performance.
Finally, we describe an infrastructural application of CAIDA's macroscopic topology mapping project (skitter), which focuses on assessing the optimality of root DNS server placement.