"Data Mining Meets Network Management: The NEMESIS Project"
Minos Garofalakis and
Proceedings of DMKD'2001,
Santa Barbara, California, May 2001.
Modern communication networks generate large amounts of operational
data, including traffic and utilization statistics and alarm/fault data
at various levels of detail.
These massive collections of network-management data can grow
in the order of several Terabytes per year, and typically hide "knowledge"
that is crucial to some of the key tasks involved in effectively managing
a communication network (e.g., capacity planning and traffic engineering).
In this short paper, we provide an overview of some of our recent and
ongoing work in the context of the NEMESIS project at Bell Laboratories
that aims to develop novel data warehousing and mining technology for the
effective storage, exploration, and analysis of massive network-management
We first give some highlights of our work on
Model-Based Semantic Compression (MBSC), a novel data-compression
framework that takes advantage of attribute semantics and data-mining
models to perform lossy compression of massive network-data tables.
We discuss the architecture and some of the key algorithms
underlying SPARTAN, a model-based semantic compression system
that exploits predictive data correlations and prescribed
error tolerances for individual attributes to construct concise
and accurate Classification and Regression Tree (CaRT) models
for entire columns of a table.
We also summarize some of our ongoing work on warehousing and analyzing
network-fault data and discuss our vision of how data-mining techniques
can be employed to help automate and improve fault-management in modern
More specifically, we describe the two key components of modern fault-management
architectures, namely the event-correlation and the root-cause analysis
engines, and propose the use of mining ideas for the automated inference and
maintenance of the models that lie at the core of these components based on
warehoused network data.