Newsletters

Server Consolidation Using TeamQuest Software
NCR was challenged with the task of reducing the number of Windows servers without jeopardizing the service levels for applications. The overall objective was to:

• Address underutilized hardware
• Leverage up-to-date processors
• Lower server hardware support costs
• Cut software maintenance costs
• Reduce physical space, power and cooling requirements

For NCR, the best way to achieve this was to integrate VMware into the environment. VMware provides an environment where multiple virtual machines (VMs) can run independently within the same physical server. At NCR, 14 physical servers have been transformed into 180 virtual servers.

Using consistent terminology within any IT group is always helpful. The term partition, for example, can mean different things to different people. It might be a slice of disk or an unconfigured part of a server, or something else entirely. Therefore, the term definitions below will be used in this article:

• Physical container - the physical VMware server
• Virtual server - the resident guest systems
• Service console - the host ESX system

With VMware, various areas need to be watched - the health of the overall container system, individual virtual servers and the service console. In addition, pay attention to how the various virtual servers are interacting with each other on the container.

"I have found TeamQuest to be a life saver in all of these areas," said Cheryl Wenner, systems capacity planner at NCR Corporation. She focuses her efforts on applying linear trend analysis to the measurement of free physical memory and aggregate CPU usage on the Container.

Linear Trend Analysis -- Memory
TeamQuest is invaluable in managing memory in any VMware environment. It can show on one screen, for example, the total amount of physical memory that exists, the amount of managed memory, the amount of memory configured among all virtual servers, and the amount of memory being used by VMs.

Interestingly, the amount of allocated memory can sometimes actually exceed the amount physically available (i.e. a physical container has 16 GB RAM and has 18 virtual servers, each of which has 1 GB of memory allocated). However, in the reality, the amount being used is what counts and requires the most management. In such a configuration, it would be unusual for all memory to be utilized by all VMs simultaneously because of VMware's memory management and sharing capabilities.

So capacity management becomes a case of finding the right memory threshold to maintain ESX performance. At NCR, a minimum of 2 GB free RAM is required. Below that, VM performance begins to suffer.

To identify that break point, TeamQuest linear trend analysis takes selected parameter values, calculates a trend line for the selected parameter, and then projects that trend line into the future.

The capacity planner must identify which parameter to trend, how much historical data to use when calculating the trend line, how far into the future to extend the line, and what threshold value to use for the trend line.

It is also possible to generate an upper and lower confidence interval. These confidence intervals help identify the range of values possible in the trend line calculation.

Here are the steps needed to set up a VMware linear trend policy for memory:

• Open TeamQuest Manager for the host
• Go to Analysis Tools - Linear Trend Policy
• Add a Definition
• Select the statistic Class - VMware - and Subclass - Summary:Memory
• Select the statistic - free_mem
• Set the Policy Name - free_mem
• Set the Time Backward - number of days to go back when calculating the trend line
• Set the Time Forward - number of days to project the trend line forward
• Set the Threshold

Free mem is already available as a statistic within TeamQuest.

For linear trend analysis, 30 days is generally enough history for accurate trending. But you don’t have to wait that long to be able to view valuable information. To see the statistics immediately, edit the linear trend analysis schedule and run the Aggregation Service from TeamQuest Manager. After opening View, select the new statistics, and then set the calendar for the 1-hour projection aggregation set. The new reports are now available.

Health of the Overall Container System - CPU
When viewing CPU usage, there are fewer layers to monitor compared to memory. TeamQuest is very good at observing, for example, cumulative CPU usage of each virtual server on the container. It also shows the usage rates for each individual VM. While the RAM levels at NCR have to be kept above 2 GB free, CPU rates should be kept below 80 percent busy.

The steps to set up a VMware Linear Trend policy for CPU are:

• Open TeamQuest Manager for the host
• Go to Analysis Tools - Linear Trend Policy
• Add a Definition
• Select the statistic Class - User Defined - and Subclass – (blank)
• Select the statistic - vm-total-cpu
• Set the Policy Name - vm-total-cpu
• Set the Time Backward - number of days to go back when calculating the trend line
• Set the Time Forward - number of days to project the trend line forward
• Set the Threshold

Alarm definitions can be set to work with linear trend analysis. If the trend line, or one of the confidence interval lines, exceeds the threshold value, an alarm can be generated. This is particularly important for some metrics that can’t wait a week to show up. Just select the appropriate alarm conditions to allow enough time to react.

Steps to set up a VMware Linear Trend alarm:

• Open TeamQuest Manager for the host
• Go to Alarm Policies - Aliases, and define an alias
• Go to Alarm Policies - Actions, and define an action
• Add the alarm

Linear trend analysis is only one option for statistical analysis of TeamQuest data. Other measures include short term analysis and minimum-maximum comparisons. Implementing any of these in a production environment can greatly enhance the proactive monitoring value that TeamQuest provides.