I am attempting for the first time to load test one of the application using Ajax true client. We plan to load with 1000-concurrent users for few business process.
Has anyone tried high vuload testing using Ajax true client? Do you have any suggestions in terms of the hardware you used for Load generators?
I was informed that Ajax trueclient are memory & CPU intensive. I will be using high end hardware, so getting few 8GB memory RAM is not a big deal...But I wonder how the CPU utilization can be controller?
I did run for 25 vu load and peak CPU utilization was 19%, I am struck on how to right size my LG's for handling 1000 users.
Thanks for your time and appreciate your thoughts.
I tried load testing an application using Ajax Truclient. My plan was to load test the application with 1000 concurrent users for some business process
After reading the Hp expert's reply , I ran the BP with 1Vuser for 1 Iteration and the average CPU utilization was 18%. Since the utilization was immensely high, I tried running it again to check whether the results are correct or not, again it showed almost the same result.
The Memory (Private working set) usage was a whopping 1405401098.952bytes.
Please give me a suggestion for controlling the CPU utilization.
We also explored the use of TruClient (both FireFox and IE) for performance testing purposes, but found similar result that you have experienced. The CPU and Memory footprint required per VUser made the protocols unusable for any real-world performance testing.
Exact quote excerpt from my written note to HP on the subject:
"...Additionally, we observed that our load generator servers were requiring to run vusers as dedicated processes (versus threads), and that each mdrv process would consume 300-500Mb of memory. With that sort of demand on the servers, it would be impractical to pursue the use of TruClient for achieving any significant load..."
We personally have not gone so far as to use another tool, since in my personal opinion LoadRunner is by far the best platform available for performance testing.
Instead, we adjust estimates for script development accordingly and continue to use the Web (HTTP/HTML) protocol for primary side load placement. Correlation can be much more complex on some of our newer internal application platforms which use extensive json/comet calls, but the end product can still be made to function predicatably and reliably.
Then, to capture any client-side metrics, we can still use a very small implementation of a couple GUI-type virtual users such as TruClient.