NeTS-NR ROSS.Net: A Platform for Integrated Large-Scale Network Design of Experiments and Simulation

Rensselaer Polytechnic Institute, Troy NY

Investigators

Abstract

National Science Foundation NETS - Research in Network Technologies and Systems CISE/CNS ABSTRACT Proposal Number: 0435259 Principal Investigator: Kalyanaraman, Shivkumar Institution: Rensselaer Polytechnic Institute Proposal Title: NeTS-NR ROSS.Net: A Platform for Integrated Large-Scale Net-work Design of Experiments and Simulation Performance analysis tools to aid design, analysis, understanding and operation of networking protocols on large-scale and in heterogeneous deployment con-texts are urgently needed in networking research. Though there are popular tools for small-scale analysis and emerging test-bed facilities, large-scale analy-sis requires tools such as scalable simulators, experiment design and empirical modeling engines (a.k.a "meta-simulation" tools). The overall goal of this project is to provide tools that significantly improve the way networking community de-signs, tunes, and studies network protocols, especially for target deployment in large-scale, heterogeneous and time-varying conditions. In particular, the pro-ject develops an experimentation platform, called "ROSS.Net", with meta-simulation and simulation capabilities. In order to rapidly interpret system be-havior, ROSS.Net employs large-scale experiment design tools that allow model-ing and optimization of protocol response, i.e. a response surface in a large-dimensional parameter space. The goal is to generate a unified search, optimi-zation and sparse empirical modeling framework with demonstrated ability to pose meaningful large-scale design questions and provide "good" models rapidly for on-line operation. For a researcher to selectively compose a set of experi-ments resulting in accurate answers to the questions posed by the model de-signer; ROSS.Net uses reversible computation techniques to enable memory-efficient optimistic parallel simulation and model construction. The work has immediate impact in applications broadly in the area of computer systems where large-scale design is crucial (e.g. operating systems, distributed systems). Particularly, features like fast-approximation and sparse modeling have poten-tial impact in areas far from computer networking such as include industrial quality control, agriculture, and bioinformatics. Dr. Admela Jukan Program Director, CISE/CNS Aug 5, 2004.