The histogram persistency service is responsible for storing histograms and N-tuples. The demands to the persistent histogram store are not much different than for other stores - except that this service will from individual physicists be rather used to write data rather than read data.
Writing N-tuples essential for physicists to refine data sets is very similar to writing event data. N-tuples are different from histograms: N-tuples read and analysed later must support navigation to source data and object linking. N-tuple entries without this functionality are of limited use.
Histograms usually can easily be transformed to flat floating arrays and hence be treated as any other variable size data type. For this reason no extra demands arise from the persistent representation of a histogram. The main difference results from the storage area: histograms will be produced and stored by many users - the results should be stored locally to optimize later access to the results.