HDF5 global heap
The HDF5 global heap stores file objects, including variable-length payloads, that datasets access through heap identifiers.
A file can contain multiple global-heap collections. They are reached through references rather than a single directory of all collections; "global" does not mean one contiguous region. Variable-length elements in dataset storage contain a sequence length and heap identifier instead of the complete payload. [1]
This is an on-disk structure, not the process-memory heap. It is also distinct from HDF5 fractal heaps, whose uses and filtering rules must not be assumed to apply to global heaps. [1]
Dataset filters do not process the separate variable-length payload heaps. A compressed chunk can therefore hold compressed descriptors while leaving referenced strings uncompressed. Raw chunk storage is insufficient to budget the complete file; the linked experiment compares three representations with byte-exact readback. [2] [3]