The tileDB paper has many details on the design and benchmark of HDF5 (including Parallel HDF5). ... We should perhaps considering low (C/C++) level interaction with TileDB to achieve high performance.
If tileDB is indeed as efficient as it claims, we should have a deeper look at genomeDB and see if it provides good performance. We could alternatively implement our own model on top of tileDB if genomeDB is not designed for variant/association analysis.
The tileDB paper has many details on the design and benchmark of HDF5 (including Parallel HDF5). ... We should perhaps considering low (C/C++) level interaction with TileDB to achieve high performance.
If tileDB is indeed as efficient as it claims, we should have a deeper look at genomeDB and see if it provides good performance. We could alternatively implement our own model on top of tileDB if genomeDB is not designed for variant/association analysis.