Hi Patrik,
When a large volume of sequential data is being processed the difference between a CPU bound workload and an I/O bound workload is determined by the performance and throughput of the supporting I/O subsystem. Actual throughput becomes a function of how fast input buffers can be replenished and output buffers emptied. While more available memory can be used to greatly increase buffers to smooth out the CPU processing ¨C I/O ?overlap sooner or later if there are no buffers available then throughput is determined by the I/O subsystem performance.
Of course, additional memory can be allocated to use larger data blocksizes and DASD full track blocking. In my work with the OS/360 Sort/Merge program I compared the use of 2341 DASD with a data block size and track size of approximately 7K to a half track and a data blocksize of approximately 28K on 3390 DASD. The I/O count dropped by a factor of 4 and so did the elapsed time. With the 2314 example approximately 84K was allocated to buffers. With the half track 3390 example 340K was allocated to buffers.
For non sequential processing additional memory can be utilized to hold an increased number and levels of indexes reducing index I/O operations. Pretty much any program can be changed from being I/O bound to CPU bound by the use of an SSD for data storage. In summary, additional region size is only of benefit if the processing program is able to effectively use it to reduce I/O operations.
Regards
Tom
