I am using JMeter to load test a TCP server.  Every time I run a test, some
large portion (like half) of my threads are timing out.  The very strange
thing is that the threads that time out (like all the other threads) have
received the correct response data.  And according to the server output they
received this correct response data almost very quickly (like 300 ms), no
where near the timeout.  What seems to be happening is all the threads are
receiving the correct data very quickly, as desired, but for some reason, a
large portion of the threads aren't closing as soon as they've received
their response. They are just sitting there, failing to close.  And so they
timeout and jmeter closes them, and yet JMeter realizes they have succeeded
so it gives them success value of "True". Yet it records their response time
as being a little over the timeout value (10,000 or 20,000 ms, whatever i
set it to... even though, as i said earlier, they are receiving their
correct repsonse after like 300 ms).  Does anyone know why this might be
happening? We are not having this problem with our JDBC Sampler.


We are using a Java TCP socket, which we create like this:

     Server Socket server = new ServerSocket(port);
     Socket s=server.accept();

And the way we write out to the socket in response to request from JMeter
client threads looks like this:

     BufferedOutputStream bos = new
BufferedOutputStream(s.getOutputStream());
     bos.write("test response data".getBytes());
     bos.flush();


We tried adding more ramp-up time but that didn't help at all. Still about
half timed out for no apparent reason.

Any help appreciated.

We are trying to do performance analysis for school but we can't use our
data for average response time or throughput or anything if half the
response time values are just roughly equal to the timeout and thus way, way
off.
-- 
View this message in context: 
http://www.nabble.com/Large-Portion-of-TCP-Threads-Timing-Out-Despite-Quick-Receipt-of-Correct-Data-tp24578177p24578177.html
Sent from the JMeter - User mailing list archive at Nabble.com.


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to