Here is my physical server configuration:
1 mongos router server (quad 2.5GHz / 16GB RAM)
2 config servers (quad 2.5GHz / 16GB RAM)
12 mongod servers (quad 2.5GHz / 32GB RAM)
each mongod server has a single instance of mongod.exe running and each of the 12 is paired as a replica set, so I have 6 replica sets each holding a single shard. I think.
(Yes, I know the recommended installation is a 3-node replica set, but this is not a production environment so a 2-node will suffice for me for now.)
Here is the code. It is running on a standalone server in a different subnet than the MongoDB instances (~50 ms away). The workflow is straightforward: (a single-threaded DOS script) in the list of data csv files, copy them one at a time locally, import it into MongoDB using mongoimport.exe, delete the local file, grab next file.
There are 105 files, each roughly 2.2G in size for a total of 233 gigs of data in 5.6 billion rows.
FOR /F %%G IN ('dir \\<server>\f$\export\*.csv /b') DO (
copy \<server>\f$\export\%%G C:\Scripts\MongoDB\daily\
mongoimport.exe --db <database> --collection users_daily --file %%G --type csv --headerline --host=<router> --port=27017
erase C:\Scripts\MongoDB\daily\%%G
)