Question about batch jobs input

Hello,

In MR, Spark, we are assuming each mapper or reducer handles portion of data. The data size for each map or reduce instance is at most equal to hdfs block size. So the input size can be estimated by the multipication of of blocksize*number of instances. Is the same assumption hold with Alibaba trace ?

Thank you for releasing these traces. The DAG information is so valuable. It would be great if it could include, the inputs to each job and the input sizes.