DFSMS supports improved performance for large, physical sequential data sets accessed using QSAM or BSAM access methods.

Data sets that are accessed sequentially can benefit from dynamic cache management; however, improved performance can be more effectively realized through the use of larger block and buffer sizes and parallel I/O processing.

Sequential data striping can be used for physical sequential data sets that cause I/O bottlenecks for critical applications. Sequential data striping uses extended-format sequential data sets that SMS can allocate over multiple volumes, preferably on different channel paths and control units, to improve performance. These data sets must reside on volumes that are attached to IBM® 9340 or RAMAC Array Subsystems, to IBM 3990 Storage Subsystems with the extended platform, or ESS.

Sequential data striping can reduce the processing time required for long-running batch jobs that process large, physical sequential data sets. Smaller sequential data sets can also benefit because of DFSMS's improved buffer management for QSAM and BSAM access methods for striped extended-format sequential data sets. Managing Batch Data describes how sequential data striping can be used in the batch environment.

Evaluate buffer usage in assembler language BSAM or QSAM programs to ensure that there are sufficient buffers to support the extended sequential data sets. A macro, DCBE, lets you specify options for enhanced QSAM, BSAM, and BPAM. For BSAM, the MULTACC option of DCBE lets BSAM I/O requests be run more efficiently. You can use MULTSDN to calculate a system-determined number of channel programs (NCP). If NCP is omitted, MULTSDN computes a value for NCP by multiplying the number of stripes by the number of blocks that can be stored on a track. As long as enough buffers are available, I/O is scheduled on all the stripes to provide increased performance.

When you update assembler programs to take advantage of improved buffer handling, consider taking advantage of virtual storage constraint relief with BSAM and QSAM support for 31-bit execution mode. Assembler programs using QSAM should specify the RMODE31=BUFF option on the DCBE macro and be recompiled to execute in 31-bit addressing mode. Programs using BSAM should allocate data areas above the 31-bit addressing line and be recompiled to execute in 31-bit addressing mode.

Recommendation: Use the logical backup and restore techniques for striped data sets having more than one stripe. These multi-part data sets can only be restored from physical backup copies if you enter an individual restore command for each part.

The benefit from sequential data striping must be evaluated in relationship to your ESCON®^® cache-capable 3990 storage control configuration. For each serially-attached cache-capable 3390 storage control in a storage group, up to four paths are available for concurrent I/O operations. Consequently, four stripes at most can be effectively used per storage control. Newer control units support more than four paths.

Related Reading: For more information about the DCBE macro, see z/OS DFSMS Macro Instructions for Data Sets.