Application requirements for Workload Manager configuration

The first phase of defining a configuration requires an understanding of your users and their computing needs, as well as the applications on your system, their resource needs and the requirements of your business (for example, which tasks are critical and which can be given a lower priority). Based on this understanding, you define your super classes and then your subclasses.

Setting priorities is dependent on what function WLM serves in your organization. In the case of server consolidation, you might already know the applications, the users and their resource requirements, and you might be able to skip or shorten some of the steps.

WLM allows you to classify processes by user or group, application, type, tag, or a combination of these attributes. Because WLM regulates the resource utilization among classes, system administrators should group applications and users with the same resource utilization patterns into the same classes. For instance, you might want to separate the interactive jobs that typically consume very little CPU time but require quick response time from batch-type jobs that typically are very CPU- and memory-intensive. This is the same in a database environment in which you need to separate the OLTP-type traffic from the heavy-duty queries of data mining.

This step is done using SMIT, or command-line interface. For the first few times, it is probably a good idea to use SMIT to take you through the steps of creating your first WLM configuration, including defining the super classes and setting their attributes. For the first pass, you can set up some of the attributes and leave the others at their default value. This is the same for the resource shares and limits. All these class characteristics can be dynamically modified at a later time.

You can then start WLM in passive mode, check your classification, and start reviewing the resource utilization patterns of your applications.

Verify your configuration, using the wlmcheck command or the corresponding SMIT menus. Then start WLM in passive mode on the newly defined configuration. WLM will classify all the existing processes (and all processes created from that point on) and start compiling statistics on the CPU, memory, and disk I/O utilization of the various classes. WLM will not try to regulate this resource usage.

Verify that the various processes are classified in the appropriate class as expected by the system administrator (using the -o flag of the ps command). If some of the processes are not classified as you expect, you can modify your assignment rules or set the inheritance bit for some of the classes (if you want the new processes to remain in the same class as their parent) and update WLM. You can repeat the process until you are satisfied with this first level of classification (super classes).

Running WLM in passive mode and refreshing WLM (always in passive mode) involves low risk, has low overhead operation, and can be done safely on a production system without disturbing normal system operation. To activate and refresh WLM, use the wlmcntrl command, invoked either from the command line or from SMIT.

Run WLM in passive mode to gather statistics by using the wlmstat command. The wlmstat command can be used at regular time intervals to display the per-class resource utilization as a percentage of the total resource available, for super classes). This allows you to monitor your system for extended periods of time to review the resource utilization of your main applications.