TruClient performance testing - scaling up
We are migrating our HTTP/Web scripts to TC. I have been disappointed with the amount of processor power I need to run scripts and hoping to draw on the experience of others with regard to scaling-up to reasonable transaction volumes. We are running a .Net application and transaction thoughputs are modest in the overall scheme of things (approx. one business function per second).
Spreading load over multiple LG's, turning off logging, etc, are all givens, I'm thinking more of scripting strategies and other ways to get better band for the CPU-buck.
Your thoughts & input appreciated.
Unfortunately, TC is going to consume a lot more of resources than a web one so there is no way to change it or to make scripting less consuming but here is a way to know how you can make an efficient test without exceeding the resources from your machine:
As @Tatiana_M_LR_PC indicated, TruClient cannot scale like Web (HTTP/HTML).
You can run some tweaks by reading this help page and some guideline how to calculate the amount of Load Generators required. If your application is publicly accessible you may consider using Load Generators located in the AWS.
Thank you to you both for your responses. I am aware TC is a much heavier protocol, but I have been surprised by the CPU demands - I can only run ~30 vusers on a 32-way physical (admittedly a little old) server. The link you provded Shlomi is very useful, thank you. Things like slowing down pacing and increasing think times (is this what 'Replay using recorded duration" means?) helps with LG capacity but not with achieving throughput targets, however. Also, and of perhaps even more concern, I am seeing different response times on different load generators - all I can attribute this to is the cpu capacity/architecture of the individual LG's. A Fiddler trace on each LG shows the HTTP durations are the same, so the driving of the browser (Firefox) can be the only variable factor - same scripts; same application; same runtime-settings... Unless you can suggest some other reason why this might be happening. This severely restricts our testing to a specific test config if we are to achieve repeatability of results.
Anyway, we knew the TC approach would not be without its issues, and scripts are much easier to maintain. We are considering a hybrid of TC & Web so we still get load with Web, but try to use TC for the complex and often-changing scripts.