[FLASH-USERS] performance of parallel hdf5

[Matthias Rempel]

Hi,

I have a problem with parallel hdf5 when restarting from a
checkpoint file. Reading in the data takes for ever
(meaning it didn't finish within 6h). In my case the checkpoint
file has around 24000 blocks (~1.1GB). The parallel write of the
file in the preceding run took less than 3min. I run FLASH2.5
on a linux cluster, the hdf5 version is 1.6.3. The problem
occurs somewhere in h5_read_unknowns in checkpoint_re.F90.

Has anybody encountered a similar problem?

Matthias
-- 
----------------------------------------------------------
Matthias Rempel               phone: 303-497-1522
HAO/NCAR                      fax  : 303-497-1589
3450 Mitchell Lane            email: rempel at hao.ucar.edu
Boulder, CO 80301, USA
----------------------------------------------------------