Synthetic workloads are more flexible than traces. For example, synthetic workloads do not require significant storage since they are generated on demand. However, synthetic workloads have to be elaborate, difficult to parameterize, and specific to a single environment in order to be realistic. Thekkath et al suggest the use of a technique called bootstrapping in order to extend the usefulness of traces. [TWL94]
In bootstrapping, only one trace is performed. Additional traces are generated by bootstrapping. Certain statistical guarantees with respect to the original ensures the usefulness of the generated traces. The basic idea of bootstrapping is as follows. Given an original sample of size N of a population with unknown distribution, we randomly select elements from the original sample to generate some number of new samples of size N up to NN different samples. Since elements are selected with replacement, the same element can appear multiple times in the new sample. [TWL94]