One of the things i'm a big fan of is to load test and validate a VDI environment with fictional users before going live with real users. To get a complete picture I like to get 3 points of view. End user, within the VDI and the underlying Hypervisor.
In the example given here, load tested VMware Horizon environment based on Instant Clones devided over 2 PODS both with one resource block. Each resource block consists of one vSAN Enabled cluster.
Enduser Perspective
One of the products used to validate the environment is LoginEnterprise.
In this case I am using this product to simulate user logins and start actions to mimic a basic workload. Of course this is not the only feature of this product, this is the function important for now! Something for another blog..
For the load test the components look like this.
EnterpriseManager: Control, monitor and configure the test, like how many users to login over a specific time period.
In this test 1000 users will login within a 60 minutes timeframe and will continue to run tests 60 minutes after the last sessions logged in.
The actions to perform:
The product has many predefined applications to simulate user actions.
Launchers: running on endpoints to start the sessions from. Can be virtual or physical.
Agent: within the VDI will start the applications within the session and perform the actions.
After the test it will show you some graphs from the users perspective. I guess the most important one since the user is what we are building the environment for...The user is not interested in CPU Ready or processes hogging the CPU.. they just want to use the applications with a good performance.
The basic view will show you how long it took to start the application or to perform the action within this application while the number of user's starts to rise.
VDI Perspective:
The product I'm using here is UberAgent.
This solutions depends on two components. A backend based on Splunk and the agent running within the VDI.
The agent in the VDI captures the many performance counters and uploads this data to a Splunk database. Uberagent will show you many data from verywell predefined dashboards. Breakdown of the user logon duration, network connections, latency's. process CPU usage and so... so much more. Really awesome!
Hypervisor Perspective
For the underlying hypervisor perspective I used Sexigraf. As written in a previous blog this appliance will give some really nice dashboards of the performance data collected from the defined vCenter's.
Imported data is collected like CPU Ready, storage latency's and so on.
The 3 products combined together:
Login Enterprise to start and mimic usersessions
Uberagent to capture show data from within the VDI
Sexigraf to show the data from the underlying hypervisor
An example of a specific finding.
When the test ran everything whas fine. Connections succesfull. The rise in sessions seems to have no effect on application or action runtime. So all good here.
When opening SexiGraf to see how the environment looked from the hypervisor perspective noticed something strange.. Some of the VDI's seems to flatline on 100% CpuUtilization.. thats odd....whats going on here?
So lets switch to UberAgent and find out whats going in those sessions and see wich process is claiming this CPU time:
Busted! A little misconfiguration in the AV solution (McAfee in this case) used by this customer sometimes caused the agent to hogg the CPU.
This is one example of a finding and how to combine three awesome products to detect this one!
Hope this was informative!
Usefull links:
Login Enterprise: https://www.loginvsi.com
UberAgent: Https://uberagent.com
SexiGraf: https://www.sexigraf.fr
Comentários