Details
-
Feature
-
Should have
-
None
-
Data Processing
-
-
-
2
-
2
-
0
-
Team_PANDO
-
Sprint 4
-
-
-
Description
For visibility processing we will need to read back petabytes of visibility data from the SDP buffer, and potentially hundreds of nodes in parallel. This is not what existing data access layers (e.g. casacore) were designed to do, and it is therefore expected that we will need to investigate alternative technologies.
As part of the NRAO collaboration, we are specifically considering using xarray APIs with zarr backends. This feature is to develop prototypes to investigate whether this can achieve throughput appropriate for SKA-style use cases.
This could potentially inform a number of ongoing efforts, both long term (data modelling / API design, technology selection) as well as short term (option for fixing WSClean bottlenecks / use for benchmarking prospective platforms).
Attachments
Issue Links
- mentioned in
-
Page Loading...