[CLUE-Tech] stress test for web servers / applications

[Jeffery Cann]

On Friday 17 January 2003 07:41 pm, Nathan Hokanson wrote:
> I think that it [Apache JMeter] can do what you need.  I haven't used it 
> in a while, and it looks like they have increased the functionality of it
> quite a bit since I last used it.

Dave,

Just last week, I tried to use JMeter to load test a single web server.  The 
main problem I ran into with JMeter is that it does not allow you to stress a 
server by simulating massive number of concurrent users.

In the JMeter 'Best Practices' section of the documentation, they suggest that 
you not use a lot of multiple threads for stress testing.  

On a P3 laptop with 384 MB of RAM, JMeter (well, the JVM) ran out of memory at 
20 threads.  Unfortunately, 20 threads did not even get my CPU usage on the 
server above 10%.  The JMeter docs suggest that to simulate concurrent users, 
you would need several client machines, each running 5-10 threads = 5-10 
users.  This was not feasible in my environment.  It may be in yours.

On this experience, I think JMeter is better suited for regression and 
automated acceptance testing.  It is also possible that I am confused on the 
threading issue to produce simulated concurrent users within JMeter.  If 
someone can clarify, I would appreciate it.

If you have used Load Runner or Win Runner by Mercury International, I think 
JMeter is closer to Win Runner.

I ended up using http_load, a C program and was able to easily get 100 
concurrent users to finally put a load on my web server.

http://www.acme.com/software/http_load/

HTH -
Jeff

-- 
    planet earth (tm)
http://jefferycann.com/