New fast compression filter for HDF5

As part of my work on the HDF5 for Python project (,
I've written a new compression filter for HDF5 datasets. It uses an
embedded version of the LZF library and was inspired by Francesc
Alted's high-speed LZO filter in PyTables. It offers the following:

* Much faster performance than DEFLATE (3x-5x faster compression and
2x-3x faster decompression versus DEFLATE level 1)
* Retains 50%-90% of the DEFLATE compression ratio
* No type restrictions like SZIP; will compress anything DEFLATE can handle.
* No runtime configuration required
* No library dependencies
* Works with HDF5 1.6 and 1.8, on Linux, Windows and Mac OS-X
* BSD licensed

(Performance data taken from single-precision floating point benchmarks)

I've decided to make a standalone, pure-C version of this filter
available for use in other HDF5 programs, in the hopes that others
outside the Python community might find it useful. More information
and downloads are available here:

Andrew Collette


This mailing list is for HDF software users discussion.
To subscribe to this list, send a message to
To unsubscribe, send a message to