Flow/Fluid-based Modeling Methodologies

It has been observed by a number of networking and PDES researchers alike, that much of the performance gains can be attributed to how and at what level of detail a network is modeled. To reduce the run-time complexity of sequential and parallel simulators, modelers have turned to using a ``fluid'' technique to approximate the flow between network partitions, such as Z-iterations [85], Kesidis and Walrand [63], itDecisionGuru [57], SSFNet [75], and the Rensselaer Scalable On-line Simulation Project  [100]. Our belief is that these modeling techniques complement the reverse computation techniques proposed here and can be leveraged to further reduce the execution time even without sacrificing model accuracy. In the case of the DARPA funded project at Rensselaer, a new ``domain decomposition'' approach to network modeling is currently under investigation. It requires the fast simulation of a single network domain. The research proposed here satisfies that need very well.