If you are investigating performance data interactively, generating graphs from command mode offers several distinct advantages. Once the data has been loaded into memory, you can view graphs and reports quickly, skipping around as dictated by your investigation, without causing the input data to be reread and reanalyzed. Also, if you need to produce output files in different formats, command mode is more efficient than using DCL commands.
To invoke command mode
$ ADVISE PERFORMANCE PSPA>
In command mode, you must first select the data you want to use, and then you specify the method you want to use to view it.
The graphs can be viewed in ReGIS mode on a terminal, and then written to a file in Postscript, for example, without having to reread and analyze the data. Also, you can switch between graphs and pie charts without reprocessing the data.
To select data, enter the SELECT command and specify the processing options you want to use. You can specify any or all of these processing options: ANALYSIS, PERFORMANCE, and GRAPHS. For example:
PSPA> SELECT GRAPHS /NODE=SUPPLY
This command selects data for node SUPPLY; subsequent GRAPH commands use this data to generate graphs.
The ANALYSIS option provides the results of the factory rules and optionally user rules that may fire as a result of your performance data.
The PERFORMANCE_EVALUATION option allows you to view and output sections of the Performance Evaluation Report and Tabular Report, which contain statistics about the system, including process and disk activity, and summaries.
The GRAPHS option lets you control how data is to be selected for subsequent graphing operations (including pie charts). The GRAPHS option has the following eight sub-options:
DEFAULT, generates graphs for IMAGENAMES, USERNAMES, and IO_DEVICES (and workloads).
DEFAULT is used in the absence of any specified graph processing options. ALL is equivalent to the complete list: IMAGENAMES, USERNAMES, HOTFILES, USERVOLUMES, IO_DEVICES (and workloads), and BY_NODE.
Note that when you are processing a large amount of data, each option can pose significant additional CPU and memory demands on your process.
If you specify NOALL, only system-level metrics are saved. NOALL is helpful when you need to select data as fast as possible, and retain the ability to generate graphs for the system-level metrics. The system-level metrics are always saved BY_NODE as well.
The processing options IMAGENAMES, USERNAMES, HOTFILES, IO_DEVICES (and workloads), and USERVOLUMES cause the selection process to maintain the graph statistics for each unique occurrence of an image name, user name, and so forth. The BY_NODE option causes these statistics to be maintained on both a per-node and a composite (all nodes) basis. The BY_NODE option can increase the memory demands to select the data, as a factor of the number of nodes being selected. See Appendix D for information on estimating virtual memory needs and selecting data.
After selecting the data, you can specify as many GRAPH, PIE_CHART, or REPORT commands as you wish. You can also select more data. The chapter “Using Command Mode Commands” describes the commands available in command mode.
Because Performance Manager provides so many different choices for predefined and custom graphs, you may prefer to use an interactive dialogue to make your selections.
You can request this interactive prompting from within command mode by entering the GRAPH/TYPE=PROMPT command. The Performance Manager's response is as follows:
PSPA> GRAPH/TYPE=PROMPT
Please select either 1) a predefined graph or 2) a custom graph
Choice: [1]:
Enter the graph type keyword (<cr> for list): cpu_utilization
A graph is produced. When you press Return, prompting continues.
|
Copyright © 2008 CA.
All rights reserved.
|
|