Improving Performance for Sequential Data Sets

To make the I/O operations required for a data set faster, the operating system provides a technique called chained scheduling. Chained scheduling is not a DASD option; it is built into the access method for DASD. When chained scheduling is used, the system dynamically chains several I/O operations together. A series of separate read or write operations, functioning with chained scheduling, is issued to the computing system as one continuous operation.

The I/O performance is improved by reducing both the processor time and the channel start/stop time required to transfer data to or from virtual storage. Some factors that affect performance follow:

Address space type (real or virtual)
Block size. Larger blocks are more efficient. You can get significant performance improvement by using LBI, large block interface. It allows tape blocks longer than 32 760 bytes.
BUFNO for QSAM
The number of overlapped requests for BSAM (NCP=number of channel programs) and whether the DCB points to a DCBE that has MULTACC coded
Other activity on the processor and channel
Device class (for example, DASD, tape) and type (for example, IBM 3390, 3490)
Data set type (for example, PDSE, UNIX, extended-format)
Number of stripes if extended-format.
An extended-format sequential data set can have 59 stripes. However, allocating more than four or five stripes generally does not improve performance. IBM recommends setting the number of stripes equal to the number of buffers. If your data set has too many stripes, you will waste virtual and real storage.

The system defaults to chained scheduling for non DASD, except for printers and format-U records, and for those cases in which it is not permitted.

Chained scheduling is most valuable for programs that require extensive input and output operations. Because a data set using chained scheduling can monopolize available time on a channel in a V=R region, separate channels should be assigned, if possible, when more than one data set is to be processed.