The parquet table downloads each column chunk individually. If a large proportion of the columns are used and there is a large number of row groups in the file, this implies many small downloads.
A strategy could be implemented to group the downloads of column chunks if
they are close
download parallelism is already high enough (having multiple downloads in parallel increases the total bandwidth)
The parquet table downloads each column chunk individually. If a large proportion of the columns are used and there is a large number of row groups in the file, this implies many small downloads.
A strategy could be implemented to group the downloads of column chunks if