Jmeter Address Already in use, getting error while running a 600/sec user concurrent load test

I am running a load test of 600/sec user for 2 min. At the end of test I get error in response "Address already in use.connect"

I have increased TCP Max port user to 200000. still I am getting the issue.

Total samples server are 60782.

Why i am facing this issue?

My jmeter is installed on windows server instance ec2.

Help me to resolve it.

1 answer

  • answered 2018-11-13 06:09 Dmitri T

    You cannot increase max ports number to 200 000 as port number is represented by an unsigned 16-bit integer therefore the maximum number of ports you can use is 65534.

    In fact the number will be less as some are reserved, some are already used by your operating system and other software.

    So I'm under impression that the steps you took to "increase TCP Max port" are a little bit wrong.

    It is also unclear whether it is a client (JMeter) or a server (application under test) issue.

    I would recommend increasing the load gradually and looking into Active Threads Over Time, Transactions Per Second and Response Codes Per Second charts. This way you will know what was the maximum number of users/requests per second you can produce without issues.

    If the throughput or number of users will not be sufficient you will need to find out the reason which could be either on application under test side or on JMeter side. If you are sure that everything is fine with your application under test on JMeter side you can check/implement the next points:

    1. Make sure you're following JMeter Best Practices
    2. Make sure JMeter has enough headroom to operate in terms of CPU, RAM, Network, Disk. You can use JMeter PerfMon Plugin for this.
    3. If the combination of above 2 points is not enough and you cannot conduct the necessary load - you will have to go for distributed testing and add another JMeter server so they would work in clustered mode.

    60782 samples in 2 minutes is 500 requests per second so consider checking where the bottleneck is.