COMPASS: Composable and Parallel Simulation of Internetworks

点击翻译按钮获取中文摘要

The unprecedented, exponential growth of the Internet has far outstripped the capacity of thetools needed to analyze modern networks. Simulation remains the method of choice for networkdesigners, researchers, and managers for many analysis problems. However, modern simulationtools are typically limited to analyzing tens to hundreds of network nodes, while studies ofcurrent and future networks such as the next generation Internet call for analyses of thousands tomillions. New, practical approaches to network modeling and simulation are needed to addressthis growing problem. Two frequently heard problems with simulation technology today are (1) the excessive amountof time required to construct a model for a large network, and (2) the inordinate amount ofcomputer time required to execute the resulting simulation program. This project, calledCOMPASS (COMposable and PArallel Simulation of internetworkS), will attack the firstproblem by defining and realizing an approach to enable rapid construction of network modelsby reusing model components developed by others, possibly using different simulation softwarepackages. The second problem will be attacked through execution of the model on paralleland/or distributed computers. In this regard, this project will focus on realizing a practical,flexible, approach to parallelization that can be widely exploited by the networking community.This approach, based on composing (federating) autonomous simulators executing on differentprocessors, cleanly separates the parallel and sequential portions of the simulation, allowingconcurrent optimization and refinement of each. A related problem that will be studied concernsexecution of models over a geographically distributed area in situations where modelcomponents cannot be easily co-located. To demonstrate this approach, an optimized parallel/distributed implementation of the ns2simulator will be realized as part of this project. This software will be realized on widelyavailable computing platforms and will be made freely available to the networking community incollaboration with the VINT/ns group at USC/ISI. Because ns2 is widely used in the networkingcommunity, the potential impact of this project is large. Moreover, the ability to easily reusenetwork models developed by others using different simulation packages can lead to a paradigmshift in the way modeling and simulation is routinely practiced in the networking community.Finally, simulation of multicast protocols will be used as a vehicle to evaluate and refine theparallel simulation designs and demonstrate their effectiveness. Multicast protocols have aparticularly strong need for scalable simulation tools. It is an area where new ideas andproposals are still being generated. Research in this project will focus on multicast protocols thatare of current interest to the Internet community; reliable multicast, layered video multicast anddata dissemination applications will be studied. This work will not only provide a proof ofconcept for the parallel simulation research, but will also provide an improved evaluationcapability to the networking community for solutions to problems of current interest.