This is a project funded by DARPA-ITO.

Principal Investigators:

• Shivkumar Kalyanaraman, ECSE
• Ken Vastola, ECSE
• Boleslaw Szymanski, CS

Project Summary:

The complexity, heterogeneity and speed of the Next Generation Internet (NGI) require new, scalable approaches to network management and control. We propose to develop a system of collaborative, on-line simulators which supports a suite of distributed network management and control functions.

Simulation has traditionally been a performance analysis tool used for validating models and measuring deviations from expected performance. Starting from this point, we propose a three-stage approach:

• Use on-line simulation to achieve desirable network protocol functions and demonstrate that autonomous on-line simulators at network nodes can add value to existing network protocols by operating on a larger time-scale and incorporating current and historical information.
• Extend this system of autonomous simulators into a system of collaborative simulators which communicate judiciously over the network. Note that the proposed system is different from what is commonly referred to as a ``distributed simulation'' in which a large simulation is distributed among many sites to improve performance.
• Demonstrate concrete network engineering applications of this collaborative on-line simulation facility.

The distributed, collaborative on-line simulation approach proposed here has several properties well-suited to the management of large, heterogeneous internetworks:

• Use of on-line simulations to achieve on-line parameter sensitivity analysis and tuning for improved stability and performance of distributed network management and control algorithms
• Each simulator is simple because it represents a local, parameterized view of the network, and hence can execute quickly.
• When some network nodes are temporarily cut off from the rest of the simulations, the remaining nodes will still produce reasonable simulation results based on the most recent information.

The approach to network management developed under this effort has the potential to affect the entire spectrum of network management tools and techniques. We have chosen the following protocols for initial implementation and study because we believe they are amenable to this approach and are particularly important functions which will require major innovations when networks grow by orders of magnitude:

• Advanced feedback-based traffic management.
• Stabilizing Internet routing algorithms.
• Admission control for premium network traffic classes.