Layout Showing the Major ISPs: Internet Atlas Gallery

Layout Showing the Major ISPs

Bill Cheswick, Bell Labs and Hal Burch, CMU
Lucent Technologies Bell Labs Innovations
Carnegie Mellon University (CMU) Computer Science Dept.

URL: https://www.bell-labs.com/about/history/#gref

Frequent traceroute-style path probes are used to build a tree showing the paths to most of the nets on the Internet. Once the layout is computed, the map can be colored to show geographical clues, network capacity, or IP addresses. In this example, color distinguishes different ISPs.

Visualization Thumbnail

Analysis

Data
. . . Sources
. . . Specification
. . . Aggregation
. . . Filters

Visualization
. . . Techniques
. . . Key Mappings
. . . Tools

Evaluation
. . . Strengths
. . . Issues

Credits/Contact Info
. . . Authors
. . . Contact Info

Full-Size Visualization

Back to Gallery

Additional Content

Internet Atlas Gallery

Layout Showing the Major ISPs

Bill Cheswick, Bell Labs and Hal Burch, CMU
Lucent Technologies Bell Labs Innovations
Carnegie Mellon University (CMU) Computer Science Dept.

URL: https://www.bell-labs.com/about/history/#gref

Visualization Thumbnail

About the Visualization

Visualization Techniques:
(Technique1): Graph layout treats data as a physical system and finds the set of node positions that minimizes the total energy. Any two nodes that do not share an edge are connected, via infinite strings, to a spring. If the nodes are further apart than the spring's resting length, no force is applied. If the nodes are closer than the spring's resting length, the spring is compressed and the nodes are pushed apart. Minimum total energy is found using a gradient descent algorithm.
(Technique2): Graph layout occurs one layer at a time starting with closest links, and then adding one hop at a time.
(Technique3): A decision is made whether to show all the paths or just the minimum distance spanning tree (considering only the shortest path to each destination).

Key Visualization Mappings:
(Mapping1): Color groups each ISP. (Layout can be colored in many ways: geographic region, network capacity, distance from test host, IP address, etc.)

Visualization Tools:
(Tool1) Standard Unix text filters.
(Tool2): Proprietary graph layout and mapping program.

Full-Size Visualization

Back

Internet Atlas Gallery

Layout Showing the Major ISPs

Bill Cheswick, Bell Labs and Hal Burch, CMU
Lucent Technologies Bell Labs Innovations
Carnegie Mellon University (CMU) Computer Science Dept.

URL: http://www.cs.bell-labs.com/who/ches/map/gallery/index.html

Visualization Thumbnail

Credits / Contact Info

Authors:

William R. Cheswick
Hal Burch
Steve Branigan

Paper:
"Mapping and Visualizing the Internet" Bill Cheswick, Bell Laboratories; Hal Burch, Carnegie Mellon University; Steve Branigan, Bell Laboratories. Proceedings of the USENIX Annual 2000 Technical Conference. San Diego, CA June 18-23, 2000.

Contact Info:
For more information, send email to:
ches @ bell-labs.com

Poster size maps are available at:
http://www.peacockmaps.com

Full-Size Visualization

Back

Internet Atlas Gallery

Layout Showing the Major ISPs

Bill Cheswick, Bell Labs and Hal Burch, CMU
Lucent Technologies Bell Labs Innovations
Carnegie Mellon University (CMU) Computer Science Dept.

URL: https://www.bell-labs.com/about/history/#gref

Visualization Thumbnail

About the Data

Data Sources:
(Source1): All registered target networks from Merit's routing arbiter database at ftp://ftp.merit.edu/routing.arbiter/radb/dbase/.
(Source2) Full network scan (this graph uses data from 28 June 1999). First, possible host names are generated for all possible destinations obtained from the routing arbiter database (Source1). Heuristics are used to pick an IP address that is likely to be in use. A slow host scan over time is then made until a responsive host IP address is found. Finally, a set of UDP probe packets are sent to each of these hosts, where successive packets increment TTL by 1 until a) the host is reached, b) an ICMP error is returned, or c) a hop is reached that doesn't respond to several pings.
(Source3) Daily DNS reverse PTR lookups on any new addresses discovered (approximately 5,000/day).

Data Specification:
(Data1) Path database: Simple text format containing one line per target network. First field is the target network in familiar form: 135.104.0.0/16. Remaining fields are in the form: name=[date:]value, and are used to track:

Path: a comma-separated list of IP numbers, followed by a status code unless the path successfully reached the target.
Probe Date: Date of most recent probe
Target: date of first probe and IP address of responsive host
Whiner: date and email address of person requesting that this network not be scanned

(Data2) Label Database: Simple text format. One entry per line containing an IP number, a label, and the date it was collected, separated by white space.

Data Aggregation:
(Aggregate1): Various groupings of IP addresses can be derived. For example, matching IP address octets to top-level domain can show approximate country location, or, in this example, IP address octets are matched to second-level domain to show ISP names.

Data Filters:
(Filter1): Edges may be removed.
(Filter2): Data may be reduced to show only shortest paths.

Full-Size Visualization

Back

Internet Atlas Gallery

Layout Showing the Major ISPs

Bill Cheswick, Bell Labs and Hal Burch, CMU
Lucent Technologies Bell Labs Innovations
Carnegie Mellon University (CMU) Computer Science Dept.

URL: https://www.bell-labs.com/about/history/#gref

Visualization Thumbnail

Evaluation

Strengths:
(Strength1): It is possible to see multiple groupings at the same time. (Strength2): Collected data is available for analysis over time.

Issues:
(Issue1): Graph shows IP path connectivity only, not actual traffic on physical links.
(Issue2): Graph reflects only outgoing packet paths. The incoming path is often different, as ISP interconnect agreements often divert traffic through different connections.
(Issue3): Graph doesn't include networks announced in the core routing tables but not contained in the routing arbiter database. Preliminary analysis suggests that these omissions may account for 20% of IP networks.

Full-Size Visualization

Back