Review on notebook 8, usecase reprojection #5
Description
While it is really nice to build such a notebook for performance and resource usage comparison, beware of comparing the same things:
- With rasterio, the reprojection is done band by band, which may explain the difference in memory consumption.
- Maybe it would be nice trying to measure reprojection in memory, as IO might be the bottleneck.
- With Dask, you also measure the client creation time which is probably not negligible in this simple case,
- Moreover, with a Dask client/cluster, I think writing to a TIF file involve a lot of locks (chunks are written in sequence)
- We probably don't write tiled Geotiff here.