The procedural interface to the database services is called PIF application interface (PAI, [Fasc91c]) and makes extensive use of automatic code generation to achieve platform independence and generate the individual language interfaces. The PAI is split into six layers with strict interfaces between each other.
Each layer calls only functions in the underlying layer. This mechanism leads to separate modules with distinct functionality as used by individual tools. Each layer is responsible for a unique storage concept of the whole PBF, with increasing complexity towards the upper layers. The application interface works on PBFs (intertool format); for data exchange with other hosts there is the PIF ASCII form (intersite format). To convert PIF files between these two formats there is the PIF binary file manager (see section 6.1.10), implemented as a separate PIF tool on top of the PAI.
The PAI is able to handle simulation data in three geometric and infinite nongeometric dimensions. Thus it is possible to read and write distributed attributes ranging from scalar to N-order tensor values on one- to three-dimensional grids. All PIF objects can be selectively and directly accessed with the PAI, either by handle or by name. The PAI will read only the necessary parts of a PBF into a cache avoiding performance drawbacks of common file-based systems.