After the text crunching that is performed on the data, the original document text is of little importance, and does not need to be stored on the grid any longer. This may suggest preferring the upload of data as a part of the grid job submission. For the scheduled grid jobs the resulting time overhead during the grid job submission is in any case of little importance. Nevertheless, it may be more efficient to separate completely these two tasks, and to pre-load the data into a temporarily storage on the grid. In this case the failing grid jobs would not need to reload the data with every additional re submission. Pre-loading of the data would also be more suitable if the processing application is run as a service that is than invoked without any additional uploads. It would, however, require an efficient system of data management that
insures that the remaining data residues are eventually removed, and that the storage space can reclaimed.