JMeter fails on long scenario

Asked by At

I have a scenario with 5K HTTP requests. When I start JMeter with it, JMeter simply hangs after about 170 users. I followed all the guidelines for successful stress testing (no listeners, headless, increased heap space).

I must say that some of those requests are a little big, the overall file is ~115M.

When I only take a subset of the requests (~100), the simulation works better (faster initialization of users, holds more than 170 users, etc).

My question is, first, as I understand JMeter loads the scenario tree and every threads plays it, there should not be any duplication, so what exactly causes this extensive load? and second, what can I do about it?

PS: when I view the system bottlenecks I notice both CPU and memory are at very high values on the long file, both of the metrics have low values on the shorter version. Anyone can explain?

PS2: the requests have about 7 seconds of delay between them

1 Answers

Arjun Dev On

First I need to let you know that if you are using a single system to do the load testing, the maximum your hardware or the port can handle at a time is 1 Gig of data. and your firewall(if any) would again receive/pass not more than I Gig of data. Try doing the same load test with Distributed System of load testing in Jmeter(Master-Slave-Distributed System). Even then, I don't think it would run for 4k requests(if these requests are heavy).

Best possible solution:

  1. Try Distributed system as I mentioned above.
  2. Try running the load test in Non GUI Mode- CLI
  3. Increase the ramp up time as needed.
  4. Increase the Ram of your system and allocate maximum available heap space to jmeter.
  5. Drastic change- Use 1. Blazemeter cloud or 2. Move the complete setup of your load testing to Amazon Server which is more reliable and scalable.