The success of advancing technologies critical to designing future-generation high-performance global networks and reliable distributed applications hinges on the available tools that can effectively prototype test, and analyze new ideas. The project will enable advances in the area of high-performance modeling and simulation of large-scale networks. The research includes an investigation of the fundamental technologies that enable real-time large-scale network simulations and the development of a real-time immersive network simulation environment. Real-time network simulation combines the advantages of both simulation and emulation by running simulation models that interact with the physical world. Immersive large-scale network simulation requires that the simulation not only capture important characteristics of the target global network, but also support seamless interactions with distributed applications in real time. The project is divided into three research thrusts: imprecise simulation, GPU co-simulation, and the development of the immersive network simulation environment. Imprecise simulation, extended from the imprecise computation technique originally designed for real-time systems, aims to achieve real-time performance of large-scale network simulations, by allowing the simulation to choose among models with different modeling representations and with variable computing requirements during run-time. GPU co-simulation exploits the computing resource of graphics processors, which are almost omnipresent on todays desktop computers and become more powerful than CPUs; certain numerical computations, such as the network background traffic calculation, can be offloaded to the graphics hardware, so that the CPUs can concentrate on more critical tasks for real-time simulation. An immersive network simulation environment will be developed based on the imprecise simulation and GPU co-simulation techniques, and will also include models of network protocols of alternative designs of the layered network structures.
The success of advancing technologies critical to designing future-generation high-performance global networks and reliable distributed applications hinges on the available tools that can effectively prototype, test, and analyze new ideas. Real-time network simulation combines the advantages of both simulation and emulation by running computerized models to interact with the physical world. Real-world distributed applications and network services, such as online file sharing and multimedia streaming, can run together with the network simulator that operates in real time. The result immersive network simulation environment provides an invaluable alternative to existing physical network testbeds for researchers designing future-generation networking and software infrastructures. Its ability to connect real applications not only provides realistic networking conditions for designing, developing, and testing global-scale network protocols and services, but also offers a way to validate network models under realistic traffic conditions. Immersive large-scale network simulation requires real network packets be injected into the simulation system and subject to the simulated network conditions computed as a result of both real and virtual network traffic traversing the simulated network and competing for simulated network resources. The computer simulation not only needs to capture important characteristics of the target network, but also needs to support seamless interactions among the applications in real time, as if they were running on the target network. That is, for someone operating a network, the behavior of the immersive virtual network should not be distinguishable from that of a physical network. The goal of this project is to investigate fundamental technologies that enable real-time large-scale computer network simulations and to develop an immersive network simulation environment to facilitate computer network research. Using real-time network simulation represents a paradigm shift from the traditional 'best effort' approaches of conducting experiments simply on existing network systems and on emulated infrastructures. In this area, the project makes important contributions by providing high-performance solutions to real-time large-scale network modeling and simulation. These high-performance solutions include the development of advanced parallel and distributed discrete-event simulation techniques, multi-resolution modeling techniques, and scalable emulation infrastructures for interactive simulation. Specific contributions from this project include: 1) the development of hybrid network traffic models, which support the integration of various network traffic representations at different modeling levels of abstraction in order to achieve desirable balance between accuracy and performance; 2) the investigation of effective techniques for combining hybrid network traffic models and parallel discrete-event simulation methods to enable large-scale network studies on today's parallel platforms; 3) the development of computation and memory-efficient network simulation techniques for constructing large-scale network models with realistic network routing and traffic behaviors; 4) the development of scalable emulation infrastructures to incorporate a large number of real applications in a virtual network experiment; and 5) extensive studies of various network applications to validate and demonstrate the potential use of the real-time network simulation techniques. In addition to above contributions, which were made to the primary disciplines of simulation, modeling and networking research, the project also benefits other research areas. The parallel discrete-event simulator developed from this project has been used for simulating cyber-security, cellular networks, wireless ad hoc and sensor networks, parallel file systems, and transportation systems. The real-time network simulator has been used in other NSF projects to facilitate hybrid large-scale network experimentation involving simulated, emulated and physical network entities. Research and education have been an integral part of this project. Many students were involved in this project, including one postdoctoral student, seven doctoral students, three master students, and twelve undergraduate students. They have been trained in areas of computer simulation and modeling, computer networking, and performance analysis. The network simulator from this project has been made public and shared with researchers across the globe. The simulator has also been used in several graduate and undergraduate-level courses. The project contributes to the advancement of computer networks and applications, which have a profound influence on our society and the way we go about our daily lives. The simulation tools have been used for modeling cyber-security and transportation systems, benefiting the safety and welfare of the public. Several industry research labs have also used the simulator to improving their commercial products.
