Details
-
Enabler
-
Not Assigned
-
Data Processing
-
-
- Quantify the impact of the changes for a simulated SKA dataset
- Quantify any impact on performance (memory and/or runtime)
-
2
-
2
-
3
-
Team_YANDA
-
Sprint 5
-
-
-
-
10.6
-
Stories Completed, Integrated, Outcomes Reviewed, NFRS met, Demonstrated, Satisfies Acceptance Criteria, Accepted by FO
-
-
Yandasoft
-
SPO-1002
Description
After the continuum imaging experiments it is clear that a few relatively small changes to yandasoft would be beneficial for ongoing work:
1) Get EveryBeam into the codebase (to be done in SP-1539)
2) Make the yandasoft WStack implementation more memory efficient. It should be possible to do this by resetting it in the snapshot imaging framework. It would allow us to retire some of the current, standalone w-stacking classes, reducing technical debt.
3) Reduce the size of uv and image buffers. There are two front runners here: check that any buffers that duplicate data are needed (e.g. dp transforms or reformatting for a given task); and remove zero-padding for psf oversampling from all worker process buffers, adding it only after merging to the master for preconditioning and deconvolution. The latter is also an example of the former, since images are stored as both normal equation objects during the merge and params objects during deconvolution.
This enabler will address points 2 and 3, in particular WStack modifications and removal of zero-padding from worker-node buffers.