Analyzing Login VSI Results
What is the most important aspect of running Login VSI tests? The test results, of course! You can use the Login VSI Analyzer to know the maximum capacity of your environment and find specific performance bottlenecks. This video should help you better understand, interpret and analyze your first Login VSI tests.
The Login VSI Analyzer will automatically sort out the data collected during the test. It will calculate if the target environment has reached its saturation point and if so, at how many concurrent sessions. This point is called the VSImax and indicates at how many users your environment can support before they experience performance degradation.
Let’s look at an example. On a server running VMware ESX with 150 Windows 7 desktops running Office 2010, I have executed a test to figure out the VSImax. A bit more detail: These desktops were 32bit VMs with 2 virtual CPUs and 1GB of memory.
On the horizontal axis you see the number of active users and the vertical axis shows the response times in milliseconds. This graph also indicates to me that there are zero stuck sessions. If you see a lot of stuck sessions in your Login VSI test, it typically indicates something is wrong in the environment.
In this chart, the VSImax of 148 is the most important number. This tells me the system can tolerate 148 users before any of them would start to notice lacking performance (such as slow logon time).
Let’s check out another test I have executed on the same platform. The only thing I have changed in the environment is an upgrade from Microsoft Office 2010 to Microsoft Office 2013 so I can measure the impact of that change.
Right away, I spot that the maximum amount of users I can run on the platform has dropped by about 20% from 148 users to just 112 users. The only thing I changed in the environment was the version of Microsoft Office. This clearly shows why every change needs to be validated before bringing it to production.
To find out what exactly is causing this bottleneck, I can overlay performance data from different systems like logfiles from your storage device, perfmon results, XenServer or as in this example, ESXtop logfiles. As long as it is captured in a CSV file, you can overlay it. In this case, CPU is the bottleneck.
Another interesting metric is the baseline. This number indicates the performance of the system with little or no stress. A higher baseline indicates a slower performance in general, so it’s best to keep the baseline as low as possible. The baseline is typically influenced by storage, anti-virus software, workspace managers, and so on. The baseline is also used to calculate the threshold of the VSImax (this is the baseline + 1000 milliseconds).