FastSort Manual

Sorting in Parallel
FastSort Manual429834-003
6-4
Using the Automatic Configuration
5. SORTMERGEFINISH stops the distributor-collector process.
Instead of specifying input files, you can use calls to SORTMERGESEND after the call
to SORTMERGESTART. Instead of specifying an output file, you can use calls to
SORTMERGERECEIVE after the last call to SORTMERGESEND, if any, or after the
call to SORTMERGESTART. The TAL example in Example 6-3 on page 6-23 shows
how to use procedure calls for a parallel sort run.
You can use SORT and SUBSORT DEFINEs to set up a parallel sort run. For more
information, see Section 7, Using SORT and SUBSORT DEFINEs.
Using the Automatic Configuration
The simplest way to set up a parallel sort run is to let FastSort automatically configure
the subsort processes for you as follows:
1. Specify the number of subsort processes.
2. Specify the name of an initial scratch file for each subsort process.
3. Start the run.
For scratch files, specify only the disk volume names. For optimum performance, use
scratch-file volumes whose primary disk processes (DP2) run in different processors.
FastSort creates temporary initial scratch files on the disk volumes you specify and
tries to put each subsort process in the same processor as the primary disk process for
the initial scratch file.
If you do not specify a processor for the distributor-collector process, FastSort tries to
put this process in a processor as follows:
For the interactive interface, in the same processor in which the SORT process is
running
For the programmatic interface, in the same processor in which the calling process
is running
The FastSort automatic configuration also includes:
A block size of 56 KB for each subsort scratch file
A memory size of 64 KB for the distributor-collector process and for each subsort
process
An extended memory segment for the distributor-collector process of at most
90 percent of the processors physical memory not locked down by the operating
system