Is saving a HUGE dask dataframe into parquet possible?

I have a dataframe made up of 100,000+ rows and each row has 100,000 columns, totally to 10,000,000,000 float values.

I've managed to read them in previously in a csv (tab-separated) file and I successfully read them to a 50 cores Xeon machine with 250GB RAM and try to write it out as a .parq directory as such:

The floats in huge.csv were saved as strings and it is 125GB.

import dask.dataframe as dd
filename = 'huge.csv'
df = dd.read_csv(filename, delimiter='\t', sample=500000000)
df.to_parquet('huge.parq')

It has been writing to the huge.parq for close to a week and the directory is 14GB and it seems like the process of saving .to_parquet is not going to stop any time soon.

And free -mh is showing that there's still memory left available but the time it's taking to save the .parq directory is tremendously slow:

$ free -mh
              total        used        free      shared  buff/cache   available
Mem:           251G         98G         52G         10M        101G        152G
Swap:          238G          0B        238G

The questions are:

• Given the size of the dataframe and machine, is it feasible to save the dask dataframe to a parquet file at all?

• Is it normal for dask and fastparquet to take so long to save huge dataframes?

• Is there some way to estimate the time it will take to save a parquet file?