Many programs use "work" data sets, which are alternately written and read, rewritten and reread. If a work data set is used, a checkpoint should be taken each time you have finished reading the data set and before rewriting it.

For example, a program may perform the following sequence of operations to produce different versions of data set A:

1. Write and then read back A1.
2. Write and then read back A2.
3. Write and then read back A3.

A checkpoint should be taken at the beginning of operations 2 and 3 before any rewriting of data set A takes place. If, for example, the job step is abnormally terminated while operation 2 is in progress, the job step can be restarted from the checkpoint taken at the beginning of operation 2. At this checkpoint, there is no need for the data in version A1.