Telecom Data Streams (Diane Lambert, organizer)


David A. James (Bell Labs)
Stream-based Data Analysis and Visualization of High-speed Wide-area Wireless Networks

Thursday 4:00-4:30, Fountain III

Abstract:

The increasing complexity in the design, deployment, and management of newer high-speed, wide-area wireless networks has motivated us to develop a system to capture, analyze, and visualize the signaling information that flows between all cell base stations and their mobile switching center, typically over a large metropolitan area. From these streams of messages we reconstitute the state of all active data sessions and connections in the network and their most important characteristics, such as their throughput, speed, data failure rates, as well as the quality of both transmitting and receiving radio channels, (signal strength, distance to base stations, number of concurrent channels, etc.) In turn, session and connection-level statistics can serve as raw data for building network-wide performance models.

In this presentation we briefly describe the hardware and software we designed for data collection and the challenges that these data volumes present (about 1GB/hour on many networks). We then proceed to describe the analysis and visualization tools that we have developed to understand some key network performance and reliability characteristics. We close with a list of limitations and outstanding problems we hope to address in the near future.


Pat Tendick (Avaya Labs)
Measurement and Analysis of Large-Scale Stochastic Service Systems

Thursday 4:30-5:00, Fountain III

Abstract:

Traditionally, performance analysis of stochastic service systems has focused primarily on queueing theory and simulation studies. However, the ability to collect and store massive amounts of data has made it possible to measure large service systems directly. One important application of this technology is the contact center. Contact centers handle large volumes of customer interactions through voice calls, e-mail, and other media. Contact centers are large, extremely complex stochastic service systems, with perhaps thousands servers, hundreds of queues, and highly nonhomogeneous arrivals. This talk examines methods for obtaining detailed and flexible performance measurements of contact centers and other large-scale stochastic service systems, based on streaming data.


Allan Wilks (AT&T Labs Research)
Waterworks for Stream Processing

Thursday 5:00-5:30, Fountain III

Abstract:

Over the years, Rick Becker and I have constructed a number of generic tools for dealing with large streams of data at AT&T. I will talk about several useful paradigms we have worked with (some water-related!) that have helped us to build robust, lightweight, flexible software, much of which has been deployed in some of the company's most critical applications, such as fraud detection and billing.