Reading Hdf5 files with many nodes performance

[Jochem Smit]

Hi, I've already asked this question on stackoverflow, here, but I thought I'd post it here as well.

Basically what I'm asking is how to speed up the reading of hdf5 files where I have many nodes (~20-40k). I've tried multiprocessing to have 12 processes read simultaneously but that only reduced reading time by a factor of two.

Does anyone have any suggestions to improve the file reading speed?

Thanks,

Jochem