Tandem Capacity Model (TCM) Manual

Figure 2 (page 22) shows a system-level process flow diagram for a sample application. Each
oval in the figure represents a process category. This application has four transactions: standard
(ST1), inquiry (INQ), big (BIG), and update (UPD). The application is described in the following
list:
• The server classes for each of the transaction types are ST1, INQ, BIG, and UPD.
• The terminal control process (TCP) for ST1 is ST1_TCP. For INQ, BIG, and UPD, it is ST2_TCP.
• The line handler for ST1 is ST1_SIM. For INQ, BIG, and UPD, it is ST2_SIM.
• All four transactions share the same disks. DISK is the aggregation of all the disk processes
in the system.
Figure 2 System-Level Process Flow Diagram of a Sample Application
Tuning the System
The second step in the capacity planning process is tuning the system.
Before the hardware requirements of the system can be accurately predicted, you must tune the
system software bottlenecks. You might want to use the Guardian Performance Analyzer (GPA) as
an aid for tuning your system. For more information on tuning, see the Guardian Performance
Analyzer (GPA) Manual and the Measure Reference Manual.
Collecting Performance Data
The third step in the capacity planning process is collecting performance data.
Performance data must be collected on the NonStop host system using Measure on each node on
which the application runs. The samples are summarized by MeasTCM for use with TCM.
Use Measure to take measurement and performance samples of your application. Samples should
have different transaction mixes and varying utilization, including peak periods. Be sure to collect
enough samples of your application so that TCM can accurately model your application. The
recommended number of samples is at least four to five times the number of defined transactions,
especially if you intend to use the Expert features of TCM to automatically apportion data. For
example, if your application has three transactions named Withdrawal, Deposit, and Customer
Update, you should collect at least 12 to 15 samples with different transaction mixes and at varying
utilization levels.
22 Introduction