Historical Perspectives

In the System/360 and System/370 days, uni-processors were extremely slow by today’s standards. A 370/158 was rated at exactly 1 MIP. Multi-processors and attached-processors emerged on several System/370 models; these allowed the workloads to be spread over multiple processors while appearing to run as one system. However, they were expensive. Attached processors weren’t particularly good for I/O-intensive workloads and were rather rare in commercial environments. Memory was expensive and many systems didn’t contain much of it. Depending on the model, the System/370 could have 64KB and up to 8MB of memory. A somewhat robust 370/158 may have had 2MB of storage. Input and output devices were minimal and slow. Virtual operating systems, such as OS/VS1 and VM/370, were in their infancy.

CPU performance was a critical measurement of performance. From a capacity planning perspective, processors could only grow vertically to the next model in the family when additional horsepower was needed. Occasionally, there was a higher-speed model in the model group. When the capacity of a single footprint was exceeded, an entirely new system would be added, often replicating the existing configuration. I/O was carefully balanced to compensate for the minimal number of slow channels. Performance monitoring was micro-managed.

In our less ancient 3090 and ES9000 days, we typically had multi-processors, and they were getting faster. A single processor on a 9021 was approximately 62.5 MIPS. The ES9000 could grow to 10 processors and be physically partitioned to run as two five-way systems. PR/SM appeared, so multiple system images could run on the entire footprint or on each physical side. The quantity of memory was increasing and was available in two varieties: fast central storage and slower expanded storage. Channels and I/O devices were expanding and getting faster. Virtual 31-bit operating systems were available.

Processors could grow only in number to accommodate the need for more MIPS, since there was only one speed of individual processors. Capacity for more applications was expanded by adding entirely new footprints. The number of engines per each physical system varied, so there may have been an eight-way system sitting next to a three-way system or uni-processor. The Multi- Processing (MP) effect was gaining momentum. Performance monitoring was becoming more extensive to cope with new and varied hardware and workloads.

With today’s z9 enterprise and business class systems, many processors are typical and they’re quite fast. A single processor z9 (3094-701) handles 700 MIPS on average. A fully configured z9- B54 can accommodate workloads needing up to about 17,000 MIPS in a single footprint. Specialty processors are now typical in many configurations. In addition, processor speeds can be “dialed down” for those shops needing subcapacity processor speeds. Real memory is plentiful (currently up to 512GB on a single z9). I/O devices are extensive and fast. Many channels are available (up to 1,024 on the z9) and the new FICON 4 channels bring data into the system at up to 4GB/second (compared to 3.0 and 4.5MB on the ES9000). Virtual 64-bit operating systems have existed for a while and various types of operating systems are typical in many customer environments. Workload Manager (WLM) has replaced the tedious task of defining an assortment of complex parameters for the System Resource Manager (SRM).

Reality Check

Past performance monitoring techniques were good for those historical environments. Today, there are too many:

• Physical resources to monitor

• Processors and processor types to consider

6 Pages