IBM 990 - Manual (Page 236)

224 IBM
^
zSeries 990 Technical Guide
Measurements are the most accurate source for processor capacity data. Modeling
techniques may produce reasonably accurate processor capacity data, assuming all pertinent
workload and hardware design and implementation factors are considered.
There is no reasonable way to construct a benchmark that simulates instruction paths and
storage reference patterns typical of a production workload without using actual production
software and activities.
IBM utilizes the Large Systems Performance Reference (LSPR) method to provide relative
capacity information, which takes into account processor design sensitivities to workload
type.
LSPR benchmarks are laboratory controlled tests of representative workload environments,
objectively measured and analyzed.
8.8.1 Large Systems Performance Reference (LSPR)
The IBM Large Systems Performance Reference (LSPR) method is intended to provide
comprehensive S/370™, S/390, and zSeries architecture processor capacity data across a
wide variety of operating systems, or System Control Programs (SCPs), and workload
environments.
To assure that the processor is the LSPR’s primary focus, the processor capacity data
reported assumes sufficient external resources, such as storage size, number of channels,
control units, and I/O devices, so as to prevent any significant external resource constraints.
LSPR data is based on a set of measured benchmarks and analysis, and is intended to be
used to estimate the capacity expectation for a given production workload when considering a
move to a new processor.
The average rate that processors execute instructions is quoted as Millions of Instructions
Per Second (MIPS). With today’s high-performance processors, the actual MIPS rate
achieved is extremely sensitive to the workload type being run, and its relationship to
underlying processor design. Therefore, the relative capacity of one processor to other will be
very dependent on the type of the work being run.
For this reason, IBM has chosen to provide capacity data in terms of work accomplished, or
throughput, in various operating systems and workloads environments, rather than in MIPS or
instruction execution rates.
Internal Throughput Rate (ITR) and ITR Ratio (ITRR)
LSPR uses the Internal Throughput Rate (ITR) metric to measure work done on different
workloads environments. ITR is computed as:
ITR = Units of Work / Processor Busy Time
The “Processor Busy Time” is normalized to 100% utilization. “Units of Work” are normally
expressed as jobs for batch workloads, and as transactions or commands for online
workloads.
ITR characterizes processor capacity, since it is a CPU busy time measurement. ITRs are
useful for determining the relative capacity between two processors running the exact same
workload environment. However, the absolute ITR values from one workload
cannot be
compared to those of a different workload environment.
The relative capacity between processors for a given workload is done by dividing the ITR of
one processor by the ITR of another to produce an ITR ratio, called ITRR. For example, to