Scalable network emulation

Abstract

Network emulation provides a testing environment where real network protocols and applications running on real network hosts can interact under controlled and repeatable network conditions. One approach to emulation involves simulating a network model, called the virtual network, with a real-time network simulator and providing an I/O interface that enables interaction between real hosts and the virtual network. To reflect the behavior of large networks such as the Internet it is important that the emulation environment be scalable. The virtual network must be able to scale in size and in traffic volume and the I/O interface must be able to scale in the number of real hosts and in real traffic volume. This thesis focuses on addressing scalability of the virtual network through the use of various simulation techniques. Parallel simulation techniques are employed in both shared memory and distributed memory environments. A simulation abstraction technique is also introduced that allows for the interaction of packet-based traffic flows and fluid-based traffic flows. With only parallel simulation techniques employed, real-time emulation performance of nearly 50 million packet transmissions per second is achieved on 128 processors for a network model consisting of about 20,000 nodes. With all simulation techniques employed, real-time emulation performance of nearly 500 million packet transmissions per second is achieved on 128 processors for a network model consisting of about 200,000 nodes.