From my experience it would make sense to check the exact type Teradata wants to have and mabye try to convert that early on. Candidates are Long Integers that would show up as BIGINT on a big data system, all sorts of time and date variables (think zoned time variables) which might be best transferred as STRINGS in the first place. In the example I try to translate KNIME column types to Hive; this may vary over DB systems.
Also you should check for column names with blanks and special characters which also a lot of systems still do not like.
Then you could try and use CSV (not alsways the best option, I know) but sometimes there is a way of using zipped CSV files that would reduce the amout of transfer.
Also if you data is very large it could make sense to do it in chunks that would be easier to upload any maybe use a fail/retry configuration than do it all in one step.