search

TRIQS / triqs_0.x

DEPRECATED -- This is the repository of the older versions of TRIQS
Other
11 stars 9 forks source link

Speed of HDF_Archive #49

Closed parcollet closed 11 years ago

parcollet commented 12 years ago

For intensive reading on large files, HDF_Archive is slow, much slower than h5py. This is due to multiple rereading of the keys of the file. Largely fixed in 4ad44c1e73

To be done : improve is_group and is_data. HDF_Archive is still 3*slower than direct h5py

mferrero commented 12 years ago

Commit 4ad44c1 introduced a little bug in the Archives. Indeed when create_group is called it does not put self._keys_cached=None. As a consequence, a new created group is not visible using keys() right after it's been created. Should the _keys_cached be put already at the HDF_Archive_basic_layer_h5py.py layer?

parcollet commented 12 years ago

yes, it should.

mferrero commented 12 years ago

I leave the ticket open, there is still the improvement of is_group is_data to be done