Large networks, such as the Internet, pose an ideal medium for solving computationally intensive problems, such as NP-complete problems, yet no well-scaling architecture for Internet-sized systems exists. I propose a software architectural style for large networks, based on a formal mathematical study of crystal growth that will exhibit properties of (1) discreetness (nodes on the network cannot learn the algorithm or input of the computation), (2) fault-tolerance (malicious, faulty, and unstable nodes cannot break the computation), and (3) scalability (communication among the nodes does not increase with network or problem size). I plan to evaluate the style both theoretically and empirically for these three properties.