Tandem Capacity Model (TCM) Manual

5 Creating a WA Model
This chapter describes how to create a customized Workload Apportionment (WA) model for a
new application (or set of applications). It is assumed in the following instructions that a Measure
sample has been collected on the NonStop host system, and that a performance sample has been
generated by MeasTCM, downloaded to the PC, and loaded into TCM. This chapter also includes
a sample WA model for a non- Pathway application.
For detailed information on selecting menus and menu items on the WA model screens, see
“Workload Apportionment Ribbon” (page 125).
Overview
You use the Workload Apportionment (WA) model to translate process category performance data
collected and summarized on the NonStop host system into transaction-oriented performance data
that can be measured in terms of CPU and disk consumption. To perform its work, the WA model
uses apportionment tables that specify how CPU and disk consumption should be charged to
transactions and how to compute the number of transactions.
The WA model is used to update the Capacity History database so that the capacity planner can
effectively track transaction-based performance data and check apportionment methods for
reasonableness. A model can be viewed as reasonable if CPU and disk demands per transaction
are relatively constant across many performance samples (CSV files) with varying transaction mixes
and throughput levels.
After you apportion performance data using the WA model, you can load it into the TCM
Performance model for more analysis. For more information about this process, see Chapter 6:
“Creating a Performance Model”.
Apportionment Tables
The TCM WA model uses two tables to determine how to apportion process category performance
data to transactions. These tables are contained in a file with an .AWA extension. The first table
is called the Apportionment Weights table. It indicates how to apportion the CPU and disk seconds
of process categories to transactions. The second table is called the Transaction Count table. It
specifies how to count transactions (how to compute the number of transactions in a performance
sample from the number of messages that process categories receive, send, or both).
Together, these two tables completely describe the apportionment method that TCM uses. Defining
these tables is a major step in creating a WA model.
In general, TCM makes the following assumptions when creating default apportionment tables
based on the performance data sample you have loaded:
• The transaction count for a specific transaction equals the sum of the messages received by
its servers.
• The CPU and disk seconds consumed by process categories are spread evenly (on a
per-transaction basis) across all transactions. In other words, if transactions Txn1 and Txn2
both use the TCP1 requester, the WA model assumes that they both consume the same amount
of TCP1 resources per transaction unless you indicate otherwise. This assumption applies for
nondisk categories only.
NOTE: Because Measure provides DISOPEN information, TCM accurately apportions disk activity
without requiring any additional input in the apportionment tables.
Example of an Apportionment Weights Table
The Apportionment Weights table is a matrix of process categories (rows) versus transactions
(columns). The weighting factors are relative. That is, the important information is the ratio of the
Overview 71