Extraction, Split User Exits, and Split File Time Processing › Setting up for Split File Time Processing › Allocate Split File Indexes (Recommended)

Allocate Split File Indexes (Recommended)

To maximize the power of the TIME operand on SPLIT statements, we recommend that one or more split file indexes be allocated. The TIME operand process uses the index to determine if more records can be written to the current generation of the GDG, or if a new data set needs to be allocated to the GDG. The index must be allocated before it is used.

To determine the size of the index file, you must have the following information:

• The number of SPLIT statements that will use the same index file, and the number of systems that each SPLIT statement will affect. For example, a SPLIT statement with the ALL operand will be run on all systems, while one with a SID(SYSA) operand will only take effect on a system where the SMF ID is SYSA.
• The number of SMF dump processes that will go on in a day. If MAN file switches and corresponding dumps are scheduled regularly, this is easy to determine, but if dumps are only scheduled for when a MAN file fills, this can be harder to predict.
• The number of days of data that need to be kept in the index.
• The number of intervals that are defined in the TIME operands on the SPLIT statements that reference this index.

The number of entries added to the index for a SPLIT statement is the number of TIME operand boundaries crossed, plus 1. For example, if a SPLIT statement is coded with TIME(0000,0800,1600), and the file being dumped spans from 0400 to 1200, two entries will be added to the index. The first entry will indicate that the data set with data up to 0800 is complete, and the second entry will show that there is data in a data set from 0800 to 1200, but that the data set is incomplete, meaning that the next dump may add more data to the data set. If TIMECLOSE is active for this split file as well, the entry containing the last record written to the split file may be marked as complete, and a new entry may be created for use by the next dump process.

Using this, an approximate number of entries can be calculated.

At half-track blocking, a one track data set can hold 290 entries. One cylinder can hold 4,350 entries, which should be sufficient for most data centers.

In order to make processing slightly more efficient, the INDEXLIM operand can be used on the SPLIT statements to indicate how large the index can be. It is not required, but if the parameter is not provided, the dump process assumes that the file can be added to at the end without any restriction and an x37 abend might result.

The data set attributes for the split file index are:

DSORG=PS
LRECL=192
BLKSIZE=192 or a multiple of 192
RECFM=F or FB