Scalable Network Simulation: Dynamic Aggregation

Despite the increases in processing capability, it remains impractical to conduct simulations of large-scale and high speed networks over reasonable periods of simulation time to collect statistics of interest. For this reason, many important networking problems are studied using "toy" or "reference" topologies, usually consisting of fewer than 10 network nodes. In this project, we are exploring the use of dynamic aggregation to speed the simulation of large network topologies.

Through aggregation, we can represent a portion of the network topology at a more coarse level of detail that can be simulated more efficiently. By dynamically changing between representation levels, we can maintain accuracy in the simulation while still achieving good speedup. The key is to simulate interesting parts of the network at a fine level of detail, while aggregating the background activity. A number of challenging issues must be addressed, including (1) designing multi-level representations, (2) triggering changes between different levels of detail, (3) identifying when to make a change in representation, and (4) integrating the aggregation within a simulation framework. We are working as part of the DARPA-supported project entitled S3: Scalable, Self-Organizing Simulation to address these and other issues in scalable simulation.

A related project is Modeling Topology of Large Internetworks, where we have developed models that reflect the topological characteristics of the Internet. We have also considered the significant of the differences between graphs generated with different methods. That project includes a publically available package, GT-ITM: Georgia Tech Internetwork Topology Models, for generating and evaluating graphs using a variety of methods.



Ellen Zegura
Last modified: Mon Jul 14 14:44:54 PDT 1997