An XML based framework for self-describing parallel I/O data

File I/O data is interpreted by high performance parallel/distributed applications mostly as a sequence of arbitrary bits. This leads to the situation where data is 'volatile' information that has meaning only with the respective application responsible for it; a fact that contradicts the physical storage paradigm. We claim that this situation must change and see the necessity to follow an approach - known from the database world - to interpret data as 'a model of reality'. To achieve this we add semantic knowledge to the datafiles. Thus we propose an XML based language, which provides homogeneous framework for describing data on all interpretative levels from physical representation to logical information. In our framework physical files are consequently augmented by interpretative information specified in the proposed language. This approach delivers, beside semantic knowledge on the data, the advantages of persistence and portability.