Visualizing Network Data

1
Visualizing Network Data

• Richard A. Becker et al.
• IEEE Transactions on Visualization and Computer
  Graphics
• March 1995
• Presented by Haixia Zhao

2
Focus

• Visualize the data associated with a network
  (instead of simply visualizing the structure of
  the network itself)
• A Network consists of a set of nodes and links
  with data associated with them.
• Geographical spatial layout v.s. abstract
  network. (circuit-switched network v.s. personal
  communication network)
• Direct v.s. indirect link data (link flow v.s.,
  link capacity)
• Categorical v.s. quantitative link/node data
  type. (type of link/node v.s. links capacity)
• Static v.s. dynamic data (capacity v.s. network
  flow in several time periods)

3
Challenge

• Coping with large data volumes
• Hundreds or thousands of nodes
• Thousands or tens of thousands of links
• Data from many time periods
• Overcome the map clutter problem

4
Previous data-reduction methods drawbacks

• Previous methods to reduce the amount of network
  data
• Aggregation for large numbers of links or nodes.
• Averaging for large numbers of time periods
• Thresholding exception reporting for detecting
  changes.
• Problem
• May obscure important information.

5
SeeNet

• A network data visualization tool using
• Static displays
• Link maps
• Node maps
• Matrix displays
• Interactive controls
• Parameter focusing
• Data filtering
• Animation

6
Dataset

• Telecommunication traffic among the 110 switches
  in the ATT network on Oct. 17, 1989, the day of
  the San Francisco earthquake.
• Data in focus network capacity and the trend of
  traffic flows.

7
Link maps

• Draw nodes spatially (on a map), and draw line
  segments between each pair of nodes for which
  there is data.
• To show the statistic data of a link.
• Color, thickness, etc.
• Data for both directions
• Split and use the half connected to a node to
• show the data with that node as the originating
  node.
• To reduce clutter, If a value is zero, the
  corresponding
• line segments is not drawn
• A negative data value can be shown using a
• dashed line.

8
Overload into and out of the Oakland node(coded
as segment thickness and color, using bisected
segments to show the directions)
9
Network-wide overload in the same time period
10
Node maps

• Aggregate link data at each node.
• Display node-oriented data by showing a glyph or
  a symbol such as circle or rectangle at each node
  on the map, coding the statistic values with the
  visual characteristics such as size, shape, color
  of the glyph.

11
A node map of call attempts
12
Matrix display

• Shows the data of each link of the network.
• Solves two fundamental problems encountered by
  the geographic display of network links.
• Undue visual prominence may be given to long
  lines.
• Long lines may overplot other lines

13
Network-wide overload using matrix display
14
Parameter focusing

• Each static display is determined by a group of
  display parameters as well as by the particular
  network data.
• The effectiveness of static displays heavily
  depends on how well those parameters are chosen.
  For example,
• Choose glyph size range in a node map to reduce
  overlapping.

15
Parameter focusing (cont.)

• Dynamic parameter adjustment can help the analyst
  to choose proper parameter values

16
Parameter focusing (cont.)

• Statistic choose what statistic data to display,
  such as absolute overload v.s. percentage
  overload. Transformations may also be needed
  (square-root, logarithms, etc.)
• Levels choose what data to display and what data
  to suppress, such as suppressing links with very
  low overload.
• Geography/Topology activate deactivate nodes
  and associated links in certain geographic area
  or out of the current zoom sub-region, so the
  analyst can concentrate on the active part.

17
Parameter focusing(cont.)

• Time choose what time point to display. The
  analyst can focus on the most interesting periods
  and look for changes.
• Aggregation dynamically aggregate statistic data
  over geographical regions or logical subsets of
  the network.
• Size adjust the overall size of the symbols
  drawn on the map, such as the size range of the
  rectangles in the node map. Large enough to
  convey information yet small enough to avoid
  excessive interference with other symbols.
• Color adjust the threshold statistic value upon
  which the symbols will be colored differently to
  show the difference.

18
Parameter focusing Line shortening(network-wide
overload)
19
Parameter focusing deactivating
nodes(Percentage of idle network capacity into
and out of one node near Chicago)
20
Direct Manipulation for parameter focusing in
SeeNet

• Enable the analyst to select interesting
  parameter values using direct manipulation
• Manipulate the display parameters dynamically
  while watching instant continuous visual feedback
  on the display. Good parameter focusing is
  achieved when the display shows meaningful
  information about the data.

21
Direct Manipulation - Identification

• Interactively identify nodes and links by
  touching them with the mouse w/o pressing the
  button
• Show node names, data values, etc.
• Indicate an anchor node first, then identify
  other nodes to show the the link data between the
  nodes and the anchor node.

22
Direct Manipulation - Linkmap parameter controls
3 vertical sliders line length of links, line
thickness, animation speed 2 horizontal controls
interactive color legend and time slider. The
color legend also has a double edged slider that
can be used to filter out some lines The time
slider sets the current time period
23
Direct Manipulation - Matrix Display parameter
controls

• Also use linkmaps interactive color legend and
  time slider parameter controls.
• Additionally, it has the capability to permute
  the rows and columns using a drag-and-drop action.

24
Direct Manipulation - Nodemap parameter controls

• 3 vertical sliders
• symbol size
• animation speed
• color sensitivity level.
• Controls the cutoff values for color changes.

25
Direct Manipulation - Animation

• Automatic animation
• Computer walks continuously over all the time
  periods. The animation speed is set by the
  Fast-Slow vertical slider.
• Manual animation
• By dragging the time bar forward or backward,
  with the display updating continuously

26
Direct Manipulation - Zooming and Birds-Eye

• Center-to-edge sweeping to zoom into a rectangle
  sub-region
• Maintaining a global context by providing a
  birds-eye view on the upper left corner.
• Pan to move to another sub-region.

27
Three interactions between Zoom and Links

• Left All line segments intersecting the display
  are drawn (too busy)
• Middle any line segments with at least one
  endpoint in the display are drawn
• Right only lines that both begin and end inside
  the display (none in this case) are drawn

28
Direct Manipulation - Conditioning

• In case of multiple related statistic variables,
  select an interesting range for one or more
  background variables, and set the display to show
  a foreground variable.
• The conditioning operation implement an and
  operation. It filters out all links whose
  background variables are not within the selected
  ranges, visually showing the intersection between
  the sets.

29
Direct Manipulation - Sound

• Node state changes activate deactivate
• Conveying slider values varying pitch that
  tracks the slider bars position
• Animation frame changes bell ringing to indicate
  the restart of animation.

30
Further examples

• Apply SeeNet to a variety of situations
• CICNet packet-switched data network
• An email communication network.

31
Nodemap- CICNet Internet Network Packet Flows
13 universities and research facilities. Big
circles for routers at the facilities. Small
circles show LAN attached to the routers. The
underlying map is schematic, not
geographic Statistic data is shown for each
router interface instead of a node (router)
32
Linkmap - Email communication

• ATT Bell Lab email statistics during a year
• Each node is an employee. A link shows the amount
  of email exchanged.
• Nodes are positioned so that uses exchanging
  large amount of emails are close to each other.
• Hastings in the center is the resident computer
  expert and system administrator. Newer employees
  are on the edge.

33
Linkmap - WWW Traffic
Primary connections from US to other countries.
34
Strengths Weaknesses

• Strengths
• Easy to understand
• Weaknesses
• No favorite sentence
• Redundant

35
What happened in this topic?

• Before this paper
• Bertin 1981 laid down some fundamental work of
  using both node and link representations as well
  as matrix representations.
• Fairchild et al 1988 desribed the SemNet system
  for displaying and manipulating a 3d view of a
  large network (not data on the network)
• Sarkar Brown 1994 described a fisheye
  distortion for visualizing the structure of
  sparse networks.
• Erick Wills 1993 use aggregation,
  hierarchical information, node positioning, and
  linked display for investigating large abstract
  networks with hierarchies. They use shape, color,
  and other visual characteristics coding node
  information and color, line thickness coding link
  information.
• NCSA 1991 added 3D graphics to display
  animations of Internet packet traffic with the
  network backbone raised above the network map.
• Koike 1994 described a system VOGUE to display
  communication patterns in parallel processing
  computer systems. It used nodes and links
  positioned in 3D and rendered w/ symbols, sizes,
  and colors. It allows interactive selection of
  viewpoints.

36
SeeNet3D

• Kenneth et al 1996 SeeNet3D expanded SeeNet in
  this paper, using 3D graphics
• Some screenshots

3D linkmap (geographical semantic)
37
SeeNet3D
A partially translucent arc map showing the WWW
traffic.
38
Cybernet

• Abel et al 2000 described CyberNet, a framework
  for managing networks using 3D metaphoric worlds.

39
Geographic administration tool based on the
building metaphore
40
Topology administration tool based on the
cone-tree metaphore
41
Distributed system admin. tool based on the city
metaphor
42
Network traffic characterization tool based on a
landscape metaphor
43
Computer admin tool based on the solar system
metaphor
44
Node layout (Zschech et al 2000)

• Tree layout using the radial technique in 2d and
  3d Eades Whitesides 1994

45
Node layout (Zschech et al 2000)

• Ring layout

46
Node layout (Zschech et al 2000)

• Sphere layout w/ the most important node in the
  center

47
Node layout (Zschech et al 2000)

• Hierarchical layout

48

• Xiao Milgram 1992 reviewed various
  techniques for displaying depth information,
  examined input devices used to interact with a 3D
  space, summarized some issues in 3D network
  visualization from psychological, task-related
  and implementational viewpoints, and designed a
  preliminary experimental program for evaluating
  various network visualization techniques.

49

• The End