You bring up a very interesting concern and one that I have written about in the LR\PC Blogs recently. You can read the entire article here:
http://h30499.www3.hp.com/t5/HP-LoadRunner-and-Performance/Are-the-tools-that-you-are-using-to-test-your-application/ba-p/6585000
Basically if you find that your LG's are impacting the results of your test, your test is essentially a bust. The key to preventing this is to know the footprint of each of your scripts and not just memory but also CPU. If your LG CPU ever goes over 80% you can not be certain the results are accurate.
PC automatically monitors hosts and you can find the monitors while running the load test. PC will also alert you if the CPU ever goes over 80% while the test is running. Remember though that PC only pulls measurements every 5 seconds so even if you are hovering around 80% I would be concerned.
For LoadRunner you will need to monitor the LG's manually and you will want to monitor the available MB of memory as well as CPU.
There are two and sometime three other finite resources that an LG could run out of while running a load test. Network Bandwidth, Disk IO, and for graphics based protocols GDI resources. The solution for every one of these is the same, reduce the number of virtual users running on that LG until you are not running out of whatever finite resource is being consumed. This requires testing your scripts in advance.
Finally, if you have your controller and LG be the same machine, this is strongly discouraged. While there is not direct formula for determining exactly how many less virtual users you will be able to run I would suggest that it is at least 1/2. In other words, if your LG could run 100 virtual users if it was alone, running on the same machine as the controller you may only be able to run 50 or less.
Hope this helps answer your question.
Regards,
Craig Drummond
HP SW Premier Support, Technical Account Manager - ALM Products.
