Unless the input file is very large, two or three SORTWKdd data
sets are usually sufficient. Two or three large SORTWKdd data sets
are preferable to several small data sets. Placing each SORTWKdd data
set on a separate device can improve performance.
For optimum allocation
of resources such as virtual storage, avoid specifying a large number
of work data sets unnecessarily.
A SORTWKdd data set should not be the same as the SORTIN data
set, the SORTOUT data set, any OUTFIL data set, or any other SORTWKdd
data set because this can cause lost or incorrect data or unpredictable
results.
Cylinder allocation is preferable for performance reasons. Temporary
SORTWKdd data sets allocated in tracks or blocks (without ROUND) are
readjusted to cylinders by DFSORT.
For disk work data sets, any valid ddname of the form SORTWKdd
or SORTWKd can be used (for example, SORTWK01, SORTWKC3, SORTWK2,
SORTWK#5, SORTWKA, SORTWKXY and so on). The ddnames can be in any
order. SORTWKd and SORTWK0d are not treated as duplicate ddnames (for
example, SORTWK5 and SORTWK05 will both be used if specified, as will
SORTWKQ and SORTWK0Q). If you specify more than 255 ddnames and the
Blockset technique is selected, only the first 255 ddnames are used.
If you specify more than 32 ddnames and the Blockset technique is
not selected, only the first 32 ddnames are used.
For tape work data sets, at least three SORTWKdd data sets must
be specified. The first three ddnames must be SORTWK01, SORTWK02 and
SORTWK03. Subsequent ddnames, if specified, must be in order from
SORTWK04 through SORTWK32, with no numbers skipped,
FREE=CLOSE cannot be specified.
Spool, dummy, subsystem, VSAM and z/OS UNIX file
system data sets, and z/OS UNIX files, must not be specified as work
data sets
Parameters relating to ASCII data should not be included for tape
work data sets.