Efficient simulation of TCP/IP networks characterized by non-rare events using DPR-based splitting

• Published in: 2001 IEEE Global Telecommunications Conference - Globecom Vol. 3; pp. 1734 - 1740 vol.3
• Main Authors: Akin, O., Townsend, J.K.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 2001
• Subjects: Analytical models; Bandwidth; Communication system control; IP networks; Performance analysis; Protocols; Quality of service; TCPIP; Telecommunication traffic; Traffic control
• ISBN: 0780372069; 9780780372061
• DOI: 10.1109/GLOCOM.2001.965875

To support the myriad of envisioned communication products of the future, there is a need to develop a network infrastructure that can provide larger bandwidth, with better control of quality of service (QoS). This activity requires performance analysis techniques that can accurately model packet-switched network protocols such as those based on TCP/IP. Due to the intractability of realistic network models, simulation is required. In this paper we present two techniques that can be used to speed up simulation of these networks, thus allowing simulation of more complex network models in the same amount of execution time as would otherwise be required. These techniques are modifications of a trajectory splitting simulation technique based on direct probability redistribution (DPR). Although DPR was originally developed to speed up simulation of networks characterized by rare events, here we consider performance indices that are based on non-rare events, such as networks that utilize TCP/IP. We develop the techniques and demonstrate their utility by applying them to: (1) a simple network model, and (2) a network model that includes a tandem connection of routers, tagged and background traffic, and a detailed TCP protocol. This network model includes a feedback path that is subject to the same network traffic as the forward path, thus incurring delay and loss. The techniques provide an order of magnitude speed-up factor when compared to standard simulation, thus allowing more design iterations or simulation of larger networks in the same amount of execution time.