FastSort Manual

Sorting in Parallel
FastSort Manual—429834-003
6-14
Understanding Statistics From Parallel Sorting
5. Avoid intermediate merge passes for subsort processes. Use enough extended
memory for each subsort process to make only one merge pass. For information
about how much extended memory you need for each subsort process with
different sizes of files, see Controlling Extended Memory on page 2-11.
If you do not have enough memory available in each processor, add enough subsort
processes to limit the number of merge passes to one.
Understanding Statistics From Parallel Sorting
For a parallel sort run, FastSort returns some statistics that apply only to the
distributor-collector process and other statistics that are totals for the distributor-
collector process and all subsort processes as shown in the following table:
Identifying the Causes of Errors
When an error occurs during a parallel sort run, FastSort can identify the SORTPROG
process in which the error occurred. FastSort can also tell you the name of the file that
caused an error.
If you use interactive commands to set up the parallel sort run, FastSort sends error
messages to the list file (which is usually your home terminal). If you use procedures to
set up the parallel sort run, you can retrieve error information with the
SORTERRORSUM procedure.
For example, if you specify a fully-qualified initial scratch file name for a parallel sort,
SORTPROG returns the following error messages:
*** ERROR *** A SCRATCH FILE CANNOT BE OPENED
OPERATING SYSTEM ERROR : 12
SCRATCH FILE: $DATA.SORT.SCRATCH
SORT PROCESS #2: (1,36)
Each line of the example is explained here:

The first line contains the FastSort error text.

The second line contains the file-system or NEWPROCESS error code.

The third line specifies the type and name of the file that caused the error.

The fourth line identifies the subsort process in which the error occurred and gives
the CPU number and process identification number (PIN) for the process.
FastSort Process Statistics
Distributor-Collector Process Only RECORDS, BUFFER PAGES, ELAPSED TIME,
INITIAL RUNS, I/O WAIT TIME, FIRST MERGE
ORDER, SCRATCH DISK, MERGE ORDER,
MAX RECORD SIZE, INTERMEDIATE PASSES
Distributor-Collector and Subsort
Processes
COMPARES, SCRATCH SEEKS,
NUMBER OF DUPLICATES