You should create a data class based on service level agreements. For example, all data sets having a low-level qualifier of LIST, LISTING, OUTLIST, or LINKLIST probably belong to the same data class, because they are typical work data sets having similar allocation characteristics.

Before you actually define any data classes, gather information about the common types of data sets in your installation. You also need to determine if only the data class ACS routine can assign data classes to data sets, if end users can assign data classes to data sets, or if you want a combination of these two policies. If you intend to have only the data class ACS routine assign data classes, you need to develop methods to identify the data in your installation. However, if you allow only the data class ACS routine to assign data classes, you should be aware that users need to override some data class attributes. For example, you should probably not code one data class for each possible amount of space that users need. Finally, you need to identify the space requirements for some commonly used data sets.

By gathering useful installation information, you can relieve your end users from specifying all of the allocation attributes on allocation requests. They can use the data class that most closely matches their needs and explicitly change the few attributes that are unique to their data sets.

You can create a data class with no attributes specified and use it to handle system-managed data sets that are allocated (but never opened) without all DCB attributes being specified. These are called "empty data sets." This might be a common situation with batch systems where the DSORG, for example, is specified in the program but not all allocated data sets are used for every run of the batch stream. The empty data sets then occupy space on DASD and cannot be migrated by DFSMShsm because they do not have a specified DSORG. To solve this, check in your data class ACS routine for allocations that do not specify a DSORG or a data class and assign them the blank data class. This causes the DSORG for the data set to default to physical sequential (PS), so that DFSMShsm can migrate the data set.

A data set that has been opened for output and closed without writing anything is called a "null data set."