Currently the file structure is tied in with the operating system routines fread(3) and fwrite(3), and binary files cannot be guarenteedJ.2 to be used across machines with different data types (size/floating point conventions etc.). This problem is only partially solved by using programs such as tsf(1NEMO) and rsf(1NEMO): it still requires physical data modification, transport and again modification. (UNICOS) cannot be read on a SUN, however datafiles on e.g. an Alliant, Multiflow and SUN are binary compatible because of IEEE floating point numbers and the proper size twos complement integers. Support exists now for byte-swapping, such that files on Sun and Dec can be read and written either way. However, machines like the CRAY supercomputer with it's deviating size and floating point format will have to convert their data as is exemplified in Section .
A possible solution is the way data is written to disk in a package as miriad: the layer just before the fread/fwrite packs the data in some predefined standard format (IEEE floating point and twos complement integers seem an obvious choice at this time). This causes a small overhead on some machines, and on other machines it is nothing more than a copy operation or even passing of pointers.