FastSort Manual

Optimizing Sort Performance
FastSort Manual—429834-003
9-9
Using a Partitioned Scratch File
The following syntax creates the file SCRATCH with two secondary partitions:
FUP CREATE SCRATCH, PART (1, $data3, 64, 8),
PART (2, $data4, 64, 8)
In this example, a primary file partition, SCRATCH, is created on the current node,
volume, and subvolume. Two secondary partitions, also named SCRATCH, are
created on $data3.<current-subvol-name> and $data4.<current-subvol-
name> on the current node.
You size a partitioned scratch file in the same manner as a non-partitioned scratch file.
To calculate the size of each scratch file partition, use the formula in Manually Creating
a Scratch File on page 9-2. Note that the file must be unstructured. For more
information about the CREATE command, see File Utility Program (FUP) Reference
Manual.
After you partition and create the scratch file, use one of the methods listed at the
beginning of this section to specify the file to FastSort. You can use partitioned scratch
files for both serial and parallel sort operations. Figure 9-2 shows a parallel sort run
with a 1 GB input file, three subsort processes, three partitioned scratch files and a
partitioned output file.
Figure 9-2. Partitioned Scratch Files in Parallel Sorting
Partitioned Input File Partitioned Output File
Partitioned Scratch File
Partitioned Scratch FilePartitioned Scratch File
Subsort
Process
Distributor-
Collector
Process
Subsort
Process
Subsort
Process
VST902.vsd