Why don't you group these users inside a ThreadGroup and launch the test in
a single batch?

On Tuesday, April 19, 2016, Tenghuan He <[email protected]> wrote:

> It takes the server 1 seconds inside to process the request.
> Each time 1000 threads was added to test the server.
>
> Here is mytest script
> #!/bin/bash
>
> for ((thread=1000; thread<=10000;thread+=1000)); do
>     sed -i -r -e "20s/[[:digit:]]+/"$thread"/" testplan.jmx  # this was
> used to replace the thread_nums in the jmx file
>     logfile="testplan-"$thread"threads.log"
>     ~/tenghuanhe/apache-jmeter-2.13/bin/jmeter.sh -n -t testplan.jmx -l
> $logfile
>     sleep 2
> done
>
> and here is the result, this time it stuck at 5k users
>
> Creating summariser <summary>
> Created the tree successfully using testplan.jmx
> Starting the test @ Wed Apr 20 01:33:43 CST 2016 (1461087223631)
> Waiting for possible shutdown message on port 4445
> summary =   1000 in     2s =  622.3/s Avg:     5 Min:     1 Max:    77 Err:
>     0 (0.00%)
> Tidying up ...    @ Wed Apr 20 01:33:45 CST 2016 (1461087225320)
> ... end of run
> Creating summariser <summary>
> Created the tree successfully using testplan.jmx
> Starting the test @ Wed Apr 20 01:33:48 CST 2016 (1461087228246)
> Waiting for possible shutdown message on port 4445
> summary =   2000 in     2s = 1056.5/s Avg:     4 Min:     1 Max:    86 Err:
>     0 (0.00%)
> Tidying up ...    @ Wed Apr 20 01:33:50 CST 2016 (1461087230220)
> ... end of run
> Creating summariser <summary>
> Created the tree successfully using testplan.jmx
> Starting the test @ Wed Apr 20 01:33:53 CST 2016 (1461087233145)
> Waiting for possible shutdown message on port 4445
> summary =   3000 in   2.3s = 1323.3/s Avg:     5 Min:     1 Max:   144 Err:
>     0 (0.00%)
> Tidying up ...    @ Wed Apr 20 01:33:55 CST 2016 (1461087235494)
> ... end of run
> Creating summariser <summary>
> Created the tree successfully using testplan.jmx
> Starting the test @ Wed Apr 20 01:33:58 CST 2016 (1461087238447)
> Waiting for possible shutdown message on port 4445
> summary +   2026 in   1.5s = 1377.3/s Avg:     6 Min:     1 Max:   117 Err:
>     0 (0.00%) Active: 43 Started: 2071 Finished: 2028
> summary +   1974 in   1.1s = 1839.7/s Avg:     3 Min:     1 Max:   200 Err:
>     0 (0.00%) Active: 0 Started: 4000 Finished: 4000
> summary =   4000 in     3s = 1572.3/s Avg:     5 Min:     1 Max:   200 Err:
>     0 (0.00%)
> Tidying up ...    @ Wed Apr 20 01:34:01 CST 2016 (1461087241073)
> ... end of run
> Creating summariser <summary>
> Created the tree successfully using testplan.jmx
> Starting the test @ Wed Apr 20 01:34:04 CST 2016 (1461087244020)
> Waiting for possible shutdown message on port 4445
> summary +      1 in   0.2s =    4.4/s Avg:   108 Min:   108 Max:   108 Err:
>     0 (0.00%) Active: 128 Started: 138 Finished: 10
>
>
>
> On Wed, Apr 20, 2016 at 1:19 AM, Neill Lima <[email protected]
> <javascript:;>> wrote:
>
> > 5.000 sounds pretty high to me already.
> >
> > How long this test runs?
> >
> > Even though you are running all locally, you can be topping up your NIC
> > capability. Try monitoring it with iostat or bmon before and during both
> > scenarios, 5k vs 6k users, it can give you some inside on this direction.
> >
> > On Tuesday, April 19, 2016, Tenghuan He <[email protected]
> <javascript:;>> wrote:
> >
> > > Sorry for my imcomplete information, I was using  JMeter in command
> line,
> > > and only captured the information I need in the jmx file.
> > >
> > > Some JMeter best practice says that there is a max threads limit for
> > JMeter
> > > threads on specific machine.
> > > https://wiki.apache.org/jmeter/HowManyThreads
> > >
> > > It seems that 6000 is not the limit of my machine because I have got
> the
> > > right result for all the test threads and there should be somewhere
> else
> > > wrong, is that right?
> > >
> > >
> > > On Wed, Apr 20, 2016 at 1:00 AM, Neill Lima <[email protected]
> <javascript:;>
> > > <javascript:;>> wrote:
> > >
> > > > Welcome,
> > > >
> > > > I am not with a computer now but do a quick search on how to use
> JMeter
> > > > from command line and you will find. It consumes way less resources
> and
> > > you
> > > > can still capture the results into CSV for later analysis.
> > > >
> > > > Also, disable unnecessary listeners like SPLine, "Show results in a
> > tree"
> > > > that consume a lot of resources.
> > > >
> > > > On Tuesday, April 19, 2016, Tenghuan He <[email protected]
> <javascript:;>
> > > <javascript:;>> wrote:
> > > >
> > > > > Hi Neill,
> > > > >
> > > > > Thanks for your quick response
> > > > > I had done 2 and 3
> > > > > 4 is not accessible now for me
> > > > > what do you mean by headless mode?
> > > > >
> > > > >
> > > > > On Wed, Apr 20, 2016 at 12:45 AM, Neill Lima <
> [email protected] <javascript:;>
> > > <javascript:;>
> > > > > <javascript:;>> wrote:
> > > > >
> > > > > > I would:
> > > > > >
> > > > > > 1- Run the test in headless mode during the Test execution
> > > > > > 2- Add more memory to the JMeter process (Xmx / Xms)
> > > > > > 3- Set the OS ulimit setting to Unlimited to enable more file
> > > > descriptors
> > > > > > 4- Setup another box to induce the load on the target server
> > > > > >
> > > > > > Try one step at a time, see if it helps.
> > > > > >
> > > > > > On Tuesday, April 19, 2016, Tenghuan He <[email protected]
> <javascript:;>
> > > <javascript:;>
> > > > > <javascript:;>> wrote:
> > > > > >
> > > > > > > Hi everyone
> > > > > > >
> > > > > > > I am loading test a Thrift Javaserver using JMeter on the same
> > > > machine
> > > > > > > When setting the 6000 threads ramp up in 1 second, all the
> > threads
> > > > got
> > > > > > > correct response data from the server, however the Jmeter
> process
> > > > hangs
> > > > > > up
> > > > > > > and stuck, netstat -an shows that all the connections are in
> the
> > > > > > > ESTABLISHED state.
> > > > > > > When the threads number is 5000, all things work as expected.
> > > > > > > Because I use Apache Thrift as the server, I do not have
> control
> > > over
> > > > > the
> > > > > > > socket close.
> > > > > > > Can anyone help me?
> > > > > > >
> > > > > > >
> > > > > > > Thanks and Best Regards
> > > > > > >
> > > > > > > Tenghuan He
> > > > > > >
> > > > > >
> > > > >
> > > >
> > >
> >
>

Reply via email to