Visualizing Massive Multi-Digraphs

1
Visualizing Massive Multi-Digraphs

• James Abello
• Jeffrey Korn
• Information Visualization Research
• Shannon Laboratories,
• ATT Labs-Research
• All the graphs copied from Visualizing massive
  Multi-Digraphs

2
Massive Graph Visualizer (MGV)

• Visualization and exploration system for massive
  multi-digraph navigation.
• Assumes a vertex set of the underlying digraph
  corresponds to leave sets.
• Out-of-core graph hierarchy and visual
  representation of each hierarchy slice.
• Implemented in C and Java 3D.
• Applied in geographic information systems,
  telecommunications traffic and internet data

3
Problems with data visualization

• Massive data size
• Bottlenecks
• I/O bandwidth
• Screen
• Solution
• Hierarchical graph slices

4
Traditional graph representation
Traditional nodes and edges representation of a
fully connected graph with 20 nodes
5
Hierarchical graph slice rationale(1)

• Build hierarchical multi-digraph layers on top of
  input multi-digraph.
• Each layer is obtained from coalescing disjoint
  sets of vertices at previous level
• In short, convert multi-digraph data into
  hierarchical data structure.

V sets, E sets
Root, Leaves, Height
6
Hierarchical graph slice rationale(2)

• Layer of each level is a subgraph with vertex and
  edges , so called Hierarchical Graph Slices.
• On each slice, less nodes, much less edges.

7
Handling two bottlenecks

• The original graph is in the external memory,
  tree is computed and stored in RAM. Engine needs
  to computes one slice for interface at a time
  upon request.
• Panoramic 3D display provides hierarchical and
  horizontal navigation thru all nodes and edges.no
  information lost

8
Slice View Interfaces

• MGV provides flexible interface.
• Works on adjacency representation matrix.similar
  to representation of Needle Grid.
• Handle massive data ATT call detail
  multi-digraph has 275million daily increment on
  260 million vertices.

9
Needle grid

• Edge maps into
• a little tick
• Lines weighted
• By color, length,
• width, orientation

10
Star Maps

• Rearrange matrix
• into circular histogram
• Well focused
• Detail data triggered
• By mouse

11
Multi-comb

• stack of star maps,single
• object represent aggregated view of
• millions of edges.
• 3D coordinates facilitates
• data evaluation.
• Useful for animation of data
• evolution

12
Multi-wedge

• Each wedge is the distribution spectrum of a
  state.
• 2D

13
Aggregated views

• Simply splice the segment
• to single bar
• User move the cursor
• into the bar for part
• information

14
Usability metrics

• Ease of Use Navigation
• Good First Impression
• High User Retention over Time
• High Learnability
• Lesser number of user errors

15
Conclusion on MGV

• Computational engine Java based user interface
• Engine runs at a web server, communication thru
  XML.
• Java provides fast rendering

• Hierarchical algorithm facilitates navigation on
  slice, actually integrates visualization and
  computation.
• Large class of massive data sets.

16
Questions ?and thank you!