My requirement is to hit a server request with 1000 concurrent users. When I did a sample load run with 100 concurrent threads(All threads start at the same moment), I observed that the "load time" is linearly increasing from the first sample to 100th sample. When I sniffed JMeter traffic using Fiddler, the load times are varying for most of the samples.
This variation is rising with the number of concurrent threads. For the 100 threads criteria,
- The difference between the maximum response times in both tools is 1.654 seconds.
- The difference between the minimum response times in both tools is 612 milliseconds.
JMeter is always showing higher response times. Sometimes the difference is reaching 5+ seconds.
I'm unable to pinpoint the reason behind this. Is it the fault of Fiddler or Jmeter? Difference of milliseconds is acceptable, but for many of the requests, it is multiple seconds when I increase threads above 200.
Here is what I did:
- Used "Synchronizing Timer" with 100 users and 0 milliseconds for timeout
- Used Cookie manager and Cache manager
- Did not add any listener
- Performed load test from Non-GUI mode with below command: jmeter -n -t C:\Users\Test0304.jmx -l C:\Users\Test0304_Log_21.jtl -H 127.0.0.1 -P 8888
- Captured Jmeter traffic from Fiddler
- Set JVM heap size to 6 GB (System RAM is 8 GB)
- No assertions, no graphs.
- Jmeter version: 3.1
- Fiddler Version: 18.104.22.168