Eleftherios Koutsofios, Russ Truscott

Operating global services of all kinds often requires extracting useful information from large (sometimes extremely large) data sets. Typically, original data has several components:

- transactions
- time series statistics
- geo-spatial locations
- node-link structures


Each of these has its own characteristics that require appropriate visualization and data mining methods.

The problem we find is that in practice, specialists rely on an assortment of individual, stand-alone tools that each focus on a particular task: network modeling and optimization, event filtering and correlation, statistics collection and analysis, accounting and billing. These tools are well-suited to detecting known patterns and solving previous encountered problems, but they don't usually cope well with situations where an new pattern is encountered, or a problem involves multiple layers or data sources.

Data Integration in the Visual Interface

Impact of customer event	Customer end-to-end views	Global views
Aerial and satellite images	Muliple data sources	Geopolitical map overlays
Customer video application	Large-scale network views	Retail store data

Swift-3D is a system for visually surfing datasets of hundreds of millions of items, with the full data available for answering queries down to individual records. Swift has a high-interaction visual interface constructed from 3D maps, 2D charts, tables, and network diagrams.

Swift has several key properties:
- runs on anything from Powerwalls to desktop PC clients
- data management that scales to hundreds of millions of records
- data integration does not rely on fixed schemas
- visual overlays of geospatial maps and images
- combines online streaming with historical data sets
- non-intrusive (can run from external data feeds)
- animation of time series

Swift is domain-independent and has been applied to applications as varied as the display of IP backbone configurations and customer clustering derived from retail store point-of-sale data. It is the foundation of the Hosting Element Visualizer (HEV), a standard option offered by AT&T's Internet Data Center service. It accepts data from almost external source- the main assumption is that the data is record oriented and can be parsed into fields. Swift can load data from standard relational databases through ODBC, JDBC or XML.

Current work concerns Generic Swift which generates instance-specific code automatically from metadata, so applications can be created and customized by non-programmers. We are also combining it with YOIX components for interactive exploration of network data, that combines statistical and network views.