Several observing modes of the Herschel space observatory will result in substantially large image or image cube datasets, up to several gigabytes in size. Having one or several of these data products in memory in the interactive reduction environment is not feasible on the computer systems available to the average Herschel observer. Therefore we have developed sliced image and cube datasets. Goal is to give the user interactive access to large images or cubes stored on disk, by adequately slicing the dataset in sub-images or subcubes that are manageable in memory.

We discuss our class modeling and java implementation of sliced images and cubes using the product pool infrastructure of the Herschel DP system. This infrastructure allows us to write products to local disk, remote servers or databases. We show how our model hides the saving and loading of slices.

We also elaborate on the slice sizing scheme we have adopted. The choice of size and dimension of slices is crucial for an optimal performance. The slice sizing strategy we present can be of use for other data analysis applications that handle large datasets.