DIMACS Workshop on Designing Networks for Manageability

November 12 - 13, 2009
University Inn and Conference Center at Rutgers, New Brunswick, NJ

University Inn and Conference Center website
178 Ryders Lane
New Brunswick, NJ 08901-8535

Organizers:

Mauricio Resende, AT&T Labs, mgcr at research.att.com

Jennifer Rexford, Princeton University, jrex at cs.princeton.edu

Network management is the continuous process of monitoring a network to detect and diagnose problems, and tuning the configuration of the underlying protocols and mechanisms to fix them. However, today's network architectures were not designed with these tasks in mind. As a result, managing data networks is, at best, a black art practiced by an increasingly overwhelmed community of network administrators. In recent years, a variety of theoretical techniques (e.g., anomaly detection, streaming algorithms, tomography, and optimization theory) have helped network administrators run their networks more effectively.

Much research on network management has (understandably) worked within the limitations of the existing networking technology. However, today's measurement data, network protocols, and router mechanisms often induce management problems that are unnecessarily difficult (or even impossible) to solve. Rather than creating new ways to retrofit network management on the existing infrastructure, this workshop focuses on the clean-slate design of network architectures with management challenges in mind from the beginning. The workshop aims to understand the limitations of supporting network management on top of today's technology, to identify the algorithmic challenges in network management, and propose new designs that induce management problems that have better, and computationally easier, solutions.

Important open questions include:

• What kind of traffic measurements should high-speed links collect to enable network administrators to observe or infer important properties of the traffic (e.g., detecting anomalies, identifying the root cause of congestion or an attack, and computing the traffic matrix)?
• What kinds of routing protocols could either adapt automatically to shifts in traffic, or have configurable parameters that are easily optimized by a network-management system?
• How can multiple, separately-administered networks or servers coordinate to detect, diagnose, and fix problems (e.g., congestion, failures, attacks, etc.) without divulging sensitive information?
• How can we design systems for monitoring traffic and performance that are robust to the presence of adversaries trying to bias the measurement results? How can a network balance load over multiple paths in the presence of adversaries trying to manipulate the system?
• How can we formally describe key network-management problems and "derive" distributed algorithms (i.e., protocols) that implicitly solve them?
• What is the right "division of labor" between the management system, the control protocols spoken between the network elements, and the local mechanisms for packet handling on each router?

To start answering these questions, this workshop will bring together experts from theoretical computer science and electrical engineering, networking and distributed systems, and network management and operations.

Next: Call for Participation

Workshop Index

DIMACS Homepage