Generally, the I/O recommendations discussed elsewhere in this section apply. In general, in this and other environments, you can use Dynamic Dataset Allocation (also known as Dynalloc or SVC99) to control the data sets you need, only when you need them, and only for as long as you need them. This adds some complexity to the application logic, but may allow the system to achieve greater through-put, since other users waiting for the same data set would need to wait for the job step to complete and release the data set allocation. If you are still using JES2 semantics, you may have to wait for END of JOB .