I really did not found answer on my question in the web. I am currently doing load tests for a web service, for example: how service will handle 15 threads in 1 seconds, for that I use Jmeter. I always get different average response times for 15 threads. When I'm in my company's inner network, I get wonderful results, but when I am at home, using lan/wifi + vpn to get access to that web services, I get horrible results. When I test it through vpn, web service can not handle 30 threads in 1 seconds, average response time is like 13 seconds, otherwise from company's network, average response time is 4-5 seconds. Also, that web service, would be also called from a system using vpn. My question is, what is correct result and correct way to test it. Test it from company's network, or though vpn?
Response time consists of the following metrics:
So my expectation is that it's not the high response time, it's more about bandwidth of your ISP and VPN connections, theoretically you can subtract these connect time and the time for the packets to travel back and forth and get the "real" response time, however a better idea would be setting up a remote JMeter slave to be "local" to the system under test and orchestrate it from your "remote" JMeter master host, this way you will be able to obtain "clean" results without these network-related slow downs.
More information: Apache JMeter Glossary