Intel AS/400 RISC Server ユーザーズマニュアル

The 2000 user comparison was done on a model i825-2473 with 6 1.1GHz POWER4 processors, 45GB
of memory, and 60 18GB disk drives configured with RAID5, in a single Domino partition. The 3800
user comparison used a single Domino partition on a model i890-0198 with 32 1.3GHz POWER4
processors.  This system had 64GB of memory and 89 18GB disk drives configured with RAID5
protection.  The 20,000 user comparison used ten Domino partitions, also on an  i890-0198 32-way
system with 1.3GHz POWER4 processors.  This particular system was equipped with 192GB of memory
and 360 18GB disk drives running with RAID5 protection.   

In addition to the test results shown above, many more measurements were performed to study the
performance characteristics of Domino 6.  One form of tests conducted are what we call “paging curves.”
To accomplish the paging curves, a steady state was achieved using the workload.  Then, over a course of
several hours, we gradually reduced the main storage available to the Domino server(s) and observed the
effect on paging rates, faulting rates, and response times.  These tests allowed us to build a performance
curve of the amount of memory available per user versus the paging rate and response time.  Based on a
paging curve study of the Domino Web Access workload on Domino 6, we determined that, similar to the
Mail and Calendaring Users workload, some additional memory was required in order to achieve the
same faulting and paging rates as with Domino 5.0.11.

The iSeries models and processor speeds described in this section are obviously dated, but the concepts
and relationships of response time and megahertz (and gigahertz) described herein are still applicable.

NOTE: When comparing models which have different processors types, such as SSTAR, POWER4 and
POWER5 it is important to use appropriate rating metrics (see Appendix C) or a sizing tool such as the
IBM Systems Workload Estimator.  The POWER4 and POWER5 processors have been designed to run at
significantly higher MHz than SSTAR processors, and the MHz on SSTAR does not compare directly to
the MHz on POWER4 or POWER5.

In general, Domino-related processing can be described as compute intensive (See Appendix C for more
discussion of compute intensive workloads).  That is, faster processors will generally provide lower
response times for Domino processing. Of course other factors besides CPU time need to be considered
when evaluating overall performance and response time, but for the CPU portion of the response time the
following applies:  faster megahertz processors will deliver better response times than an “equivalent”
total amount of megahertz which is the sum of slower processors.  For example, the 270-2423 processor
is rated at 450MHz and the 170-2409 has 2 processors rated at 255MHz; the 1-way 450MHz processor
will provide better response time than a 2-way 255MHz processor configuration.  The 540MHz, 600MHz,
and 750MHz processors perform even faster.  Figure 11.3 below depicts the response time performance
for three processor types over a range of utilizations. Actual results will vary based on the type of
workload being performed on the system.  

Using a web shopping application, we measured the following results in the lab.  In tests involving 100
web shopping users, the 2-way 170-2409 ran at 71.5% CPU utilization with 0.78 seconds average
response time. The 1-way 450MHz 270-2423 ran at 73.6% CPU with average response time of  0.63
seconds.  This shows a response time improvement of approximately 20% near 70% CPU utilization
which corresponds with the data shown in Figure 11.3.   Response times at lower CPU utilizations will
see even more improvement from faster processors.  The 270-2454 was not measured with the web

IBM i 6.1 Performance Capabilities Reference - January/April/October 2008

©

 Copyright IBM Corp. 2008