search

ATNF / yandasoft

Astronomical Calibration and Imaging Software
Other
16 stars 8 forks source link

How to speed up Linmos? #14

Closed prlahur closed 3 years ago

prlahur commented 4 years ago

Linmos takes a long time to complete. Currently it's runs parallel by distributing work for different channels. Some dataset, however, has only 1 channel. Is there another way to speed things up? OpenMP? Link to internal issue: https://jira.csiro.au/browse/AXA-627

prlahur commented 4 years ago

Ping @davepallot and @jmarvil

prlahur commented 4 years ago

@davepallot, please send me small dataset and its parset to linmos, so we can quickly compare its performance with various improvements. A low hanging fruit seems to be compiling with OpenMP option enabled. Thanks

prlahur commented 4 years ago

Built the Docker image with OpenMP enabled. Just uploaded to the unofficial account: https://hub.docker.com/r/lahur/yandasoft-mpich

prlahur commented 4 years ago

Linmos-mpi works with OpenMP, but somehow the one in Singularity is still in single thread. I'm working on it, using a small dataset from Mark Wieringa (who has implemented the code in Yandasoft)

prlahur commented 3 years ago

Mark Wieringa has improved the speed of linmos-mpi. So the version natively installed on Galaxy is faster

prlahur commented 3 years ago

Closing this old ticket, as the released version 1.1 has already addressed this problem