You wouldn't need to transfer the data in or out. You can generate the data using a map/reduce program and check it using another map/reduce cluster. It takes a

You wouldn't need to transfer the data in or out. You can generate the data using a map/reduce program and check it using another map/reduce cluster. It takes about 4-5 minutes to generate the data (it is longer because we write 3 replicas of the input data on 3 different nodes). The checker program takes a couple of minutes to make sure that all of the data is correctly sorted. There aren't any indices, just the flat text files for both input and output.