A Monte Carlo simulator for evaluating server placement within network topology designs

This paper presents a Monte Carlo simulator (MCS), which is embedded within a computer-aided design tool called iCAD. iCAD is a design and analysis tool that concurrently synthesizes and integrates network topology, server placement and file allocation for application-specific networks. We formulated the three problems as a single optimization problem, where the objective is to minimize the design cost, and to satisfy both the design and performance constraints. iCAD used an evolutionary approach to search the design space; therefore, we embedded two performance evaluation methodologies: analytical and simulation within iCAD. During each evolution cycle, an analytical queuing model (network of M/M/1 queues) is used to evaluate the performance of the network topology, which carries both the client-to-client and client-server traffic. On the other hand, a Monte Carlo simulator (MCS) is used to evaluate the performance of the placed servers within the network. MCS is used due to some uncertainty concerning with the timing of file-transfer requests by the clients that could lead to unacceptable performance by the servers. Uncertainty is introduced by randomizing the timing of files' requests by each server, and then simulating the file-transfer within a server over time. The two heterogeneous analysis methodologies (queuing and simulation) complement each other in evaluating each candidate design as fast as possible; moreover, our experimental results have demonstrated the effectiveness of iCAD in finding good designs from a large design space (65 to 150 client nodes) in a reasonable amount of time.