In multiple steps; if you want to simulate 16 Gb of data and have 8 Gb of ram you simulate 6 Gb of data , write it to disk, and then do it two more times using write.table with append = TRUE the next two times to end up with a 18 Gb object, above your own ram capacity. Sorry I cannot use any example, as I don't know which exactly type of data are you planning to use (time series, spatial data etc), You can just use classical programming to make it meaningful, changing some of the covariates each simulation step to match some kind or real-example kind of dataset.