Embodiments of this invention are directed to a system and method for characterizing and modeling a virtual synthetic file system workload. In one embodiment, a virtual synthetic system is adapted to select a first location in a prior generation dataset of a first cluster and generate a first offset using a distance distribution function. Thereafter, the virtual synthetic system selects a second location in the prior generation dataset of a second cluster, wherein the second location is offset from the first cluster by the first offset. Finally, the virtual synthetic system modifies each cluster selected on the prior generation dataset thereby creating a next generation dataset. This process is repeated to generate multiple generations of a dataset. Other embodiments are also described herein.