For large projects, one often cannot handle all relevant time-series data in RAM. This problem might already show up when loading the required input time-series but intensifies when trying to save the time-series of internal states or fluxes, which are often spatially more detailed than the input time-series.

For such cases, one can tell HydPy to handle individual input or output time-series internally on disk. The related interface is very flexible and convenient, as it offers the same access as if the data were stored in RAM. However, the technical implementation is much to slow at the moment, increasing simulation times way too much. Also, there is a limitation in the number of time-series to be handled on disk (see issue #18). Both problems are due to creating a separate binary file for the time-series of each individual sequence object.

Our idea is, to get rid of the old binary format (".bin") and to use NetCDF files (".nc") instead. NetCDF also provides random access and allows storing multiple time-series within one file. Furthermore, through using NetCDF both as our "external" and our "internal" file format, we could avoid file format transformations before and after simulation runs. (We would only need to provide a function for transforming data from one "external" data format to the other, but this is something we should implement someday anyhow.)