I am working on one scenario where I have to check login flow of application for 10KTPM in minute.
In my login flow exist- load the URL, Login and Home page load.
In entire login flow we have 55 request.
Hence for this to achieve target of 10KTPM, I have used 182 users * 55 request, could you help me is the correct way I did it?
If TPM stands for "transactions per minute" the question is what do you call a "transaction":
Depending on the answer and response time you can calculate how many virtual users you need using i.e. Little's Law
Alternatively you can go for Stress Test, i.e. start with 1 user and gradually increase the load until you reach 10KTPM or response time starts exceeding acceptable or errors start occurring, whatever comes the first.
If your load test objective is to measure the time taken by ten thousand reqests, then running 182 operations (login sequences in your case) involving 55 requests each will do that. But here’s the thing: you don’t know until you run your test how long it will take. If nobody has done this load testing to your system before, it’s extremely unlikely it will get done at that rate, the point of load testing is push your system really hard and discover bottlenecks and deadlocks. You may even get server crashes from RAM exhaustion when you start doing these tests.
If your purpose is to discover the maximum number of login operations your system can handle in a minute, that makes more sense operationally. When I’ve done this sort of thing, I’ve started with five virtual users and had each of them run the login sequence repeatedly. Run the five-user test for five minutes and see how many operations complete. That usually finds some flaws in the way the 55-request sequence is coded in jmeter.
One minute typically is not enough time to get the test ramped up and running at full tilt.
Then run a ten user test. Then 20, 50, 100. You start to need a lot of VMs to run the virtual users.
I suspect your system will crumble under these tests. (I’ve done a lot of this and systems almost always crumble when first tested this way).
If my experience is any guide, one of the ways your team will fix your system to handle a large number of concurrent login operations is to reduce that request count from 55 to something much smaller, dramatically simplifying the login sequence.
Welcome to the secret fraternity of load testers!
182 users * 55 request is good if you consider all 55 sub request in 1 transaction and 1 user runs this 1 transaction without any think time (constant timer or anything else).
