Technical Limitations of Load Generators in ver 12.0?
With ver 12 having an option to be hosted on cloud, how do you work around technical limitations of virtual load generators such as clock drift, tru client CPU memory limitations etc.? Virtual generators might affect performance results and what is the solution?
I am not sure I fully understand the question.
Why do you think it would be different than using virtual machines as load generators in your lab?
There are various posts from HP and other Loadrunner users how they had issues with using virtual load generators. Clock drift can increase or decrease response time. There are limitations on the number of Vusers you can run with TruClient on virtual machines and high CPU has been observed in some cases. OS counter information can be different than what you see when you use physical generators. With ver 12.0 on cloud, do these factors affect performance test results in anyway?
Oh. Now it is clearer.
Generally speaking, as long as system resources are not over utilized there should not be an impact on performance test results. Also the clock dritfing issue has been minimized significantly since the early days of Virtual Machines. At least this is what the VM machines vendors are stating.
I recommend monitoring the system resources and use extra machine over over utilizing existing one.
Specifically regarding TruClient it is recommended also to monitor amount of context switches and make sure there is no more than 15K context switches per core.
The above is true for both physical and virtual in house or in the cloud.
Thanks for the reply. Couple more questions:
How do you monitor and manage the vm to ensure resourses aren't over utilized. Can they automate the creation of a new vm when a certain threshold is tripped? (In otherwords an on demand capacity increases).
What is the threshold when these issues start to manifest themselves?
Could you provide us with some data from VM vendors if you have that supports clock drift is no longer an issue?
What tool is used to monitor the 15k context swithes and what is the course of action when that threshold is approached?
I'm trying to understand how do you address all the performance issues relating to VM's specifically not just the ones above? How do we make sure the results gathered from VM's are accurate and there is nothing that we missed that might have happened in the VM from a OS counter perspective?