External transaction rate is the metric typically used for comparisons of systems
that may have system resources limiting the processor's ability to do work.
The VSE/CICS workload, designated light I/O in the graph, consists of light to
moderate transactions from diverse business applications including:order entry,
stock control, inventory tracking, production specifications, banking, hotel
reservations, and teller systems.There are 17 unique
transactions using various
combinations of CICS functions.This workload issues about three I/O requests per
transaction.
The I/O intensive work reflects the I/O rate of the RAMP-C workload.RAMP-C is a
synthetic interactive workload consisting of four transaction classes differing in
complexity. Programs are written in COBOL and are designed to access indexed
and sequential files.This workload is characterized by 17 I/O requests per
transaction.
The pair of Integrated Server bars in the chart show that relative performance
decreases when I/O rate is increased.The I/O content of the work determines the
throughput rate (ETR) that can be sustained by the IBM S/390 Integrated Server
before I/O contention causes the response time to exceed the one second internal
response time limit.
Since the initial un-tuned capacity of the IBM S/390 Integrated Server is limited by
the I/O intensity of the work being done and not by processor capacity or paging
caused by adding users, I/O elimination/speedup techniques can play a role in
improving throughput.I/O response times on the IBM S/390 Integrated Server
benefit from use of the latest hard disk technology and by using a READ cache in
S/390 storage or a HPFS cache in IOSP storage.When using HPFS, handling of
write requests (write policy) may become a performance - data integrity trade-off.
Although best response time and throughput may be achieved when the cache
control policy allows Fast Write (write completion status to be given before data is
actually written to the disk storage device), this benefit is achieved while creating
an increased data integrity exposure.Environments requiring maximum data
integrity must not allow disk WRITES to be cached and accept the associated
response time and throughput performance.For
these environments, the relative
number of READ and WRITE requests per unit of time and the effectiveness of the
READ caching are key performance parameters.The data in Figure6 on page34
is based on use of a 2MB HPFS cache with forced write-through policy, a
non-cached SSA disk controller, and an I/O READ to WRITE ratio of two.
I/O elimination through the use of VSE/ESA exploitation can also improve response
time, throughput, and relative performance.Since the value of exploitation can only
be determined for specific well understood workloads and environments, its benefit
is not included in Figure6 on page34.The capacities shown in the IBM S/390
Integrated Server set of bars applies to either 370 or ESA modes as these two
modes offer comparable performance when ESA exploitation is not included.S/390
storage caches and higher read-to-write ratios increase the relative performance of
the IBM S/390 Integrated Server processor while lower read-to-write ratios and
smaller caches reduce relative performance.
VM/VSE guest environments will see guest to native ratios comparable to those
seen on ES/9000 processors (approximately 0.9 for V=R guests and around 0.7 to
0.8 for V=V guests).V=R guests with dedicated I/O migrating from 370 mode to
ESA mode will have a slightly better capacity than those shown in Figure6 on
Appendix A.Performance35