Tandem Capacity Model (TCM) Manual

Assumptions
To keep TCM simple and easy to use, TCM is based on the following assumptions:
• Your HP NonStop system is well-tuned. (You can tune your HP NonStop systems by using the
Guardian Performance Analyzer, HP product number T6942.) If the system is not well-tuned,
response times might be higher than expected, more CPUs or disks might be necessary to
support predicted workloads, or both.
• TCM computes only the CPU and disk components of response time. It does not compute
end-to-end response time, which includes communication line time.
• TCM does not model memory, disk controllers, or disk space, and it is assumed that none of
these would cause a performance problem.
• TCM is not intended as a configuration or system resource manager.
• If an HP NonStop Transaction Management Facility (TMF) audit volume exists, TCM assumes
that the audit disks do not have any application data residing on them and TMF activity will
be distributed equally among the audit volumes.
NOTE: Because of the sensitivity of the HP NonStop SQL/MP optimizer to such database attributes
as demographics and selectivity, use caution in modeling SQL applications using TCM. Access
plans might change as the application’s database grows. However, TCM accurately models SQL
applications under most conditions.
Caveat
Sometimes, if the actual and the planned systems are different processor types, the mileage might
vary depending on the situation or application. The processor speed factors are based on the
Order Entry internal benchmark. This benchmark has shown excellent correlation for a variety of
capacity planning situations for a large number of customers, but some particular situations might
scale differently.
The recommendation that TCM provides is a starting point only. Capacity planners need to analyze
their own system environment and budgetary constraints for a definitive solution.
Architecture and Components
The overall architecture of TCM appears in Figure 1 (page 20). On the NonStop host system, you
use Measure to collect performance samples on your application and then analyze and aggregate
the data provided by Measure using MeasTCM. MeasTCM provides process category data used
by TCM on the workstation and creates downloadable files that detail the process category activities
of your application.
TCM has two major components:
• Workload Apportionment (WA) model
The WA model is used to apportion the process-oriented performance data collected and
summarized on the NonStop host system through Measure and MeasTCM into
transaction-oriented CPU and disk demands. The WA model also maintains a Capacity History
database of performance information. It requires the data supplied by MeasTCM.
• Performance model
The Performance model is used to answer what-if questions regarding the transaction information
that was computed by the WA model. It enables you to pose questions about the behavior of
system performance factors, such as response time, average CPU utilization, and average
disk utilization, under increasing transaction rates.
You often have a number of options regarding the system’s response to increased workload.
For example, you can add hardware to reduce CPU utilization and improve response time,
or you can leave hardware unchanged and tolerate longer response times.
Architecture and Components 19