Hi Tim,I changed the max heapsize in the env script as followings:set ACTIVEMQ_OPTS=-Xms4G -Xmx4G -Xss1m -XX:NewSize=1G -XX:MaxNewSize=1G -XX:SurvivorRatio=8 -XX:TargetSurvivorRatio=50% -XX:MaxTenuringThreshold=2 -XX:PermSize=128m -XX:MaxPermSize=256m -XX:+HeapDumpOnOutOfMemoryError -XX:+UseBiasedLocking Now these jmsexceptions have decreased dramatically. the broker can support 1000 connections now. but I think it could be more, maybe at least four or five thousand connections since I watched from jconsole that the total consumed physical memory is merely 1GB，almost 10% CPU. Is that possible to support 4-5 thousand connections simutanously with 10Kb size message per connection using single broker? How to achieve it?I found from the activemq.log file that if I continuously load test above 1000 connections, the connection will be reset or droped. It makes me think that 1000 connections is single broker's limit. but the memory/cpu/network performed good enough. so puzzled.. where my activemq.xml configuration is different from the default one: <transportConnector name="mqtt" uri="mqtt://0.0.0.0:1883?maximumConnections=40000&wireFormat.maxFrameSize=1048576000&transport.defaultKeepAlive=3600000&wireFormat.maxInactivityDurationInitalDelay=30000&wireFormat.maxInactivityDuration=0"/> I am using Paho M2mqtt to send mqtt messages.
> Date: Sat, 22 Aug 2015 12:50:54 -0400 > From: tabish...@gmail.com > To: users@activemq.apache.org > Subject: Re: jmsexception unknown PUBREL xx received > > On 08/22/2015 12:17 PM, wangjunwei wrote: > > Hi all, I am on a project and doing stress test to activemq. I noticed lots > > of WARN messagesJmsexception unknown PUBREL xxx recived when I launched > > above 800 connections from our four different windows server simutanously. > > which xxx is an integer numbermeanwhile I noticed lots of > > jave.net.socketexception connection reset, write error and etc.I am using > > activemq 5.11.1 and M2mqtt 4.1.0.0 paho .net library as MQTT client. and > > the per mqtt message size is 10Kb. the broker configuration is the default > > setting.but from the windows performance monitor, the captured data showed > > that CPU is less than 70% on average,memory is used only 1Gb, disk write > > 40Kb/s no disk queue, network bandwidth consumed 80%. all seems good.if I > > use less than 200 connections, there is no these exceptions.can anyone > > sheld light on this? Thanks > I believe that the default max heap in the env script for the broker is > 1GB so if you hit that then you would be sure to see troubles like > this. If you are going to stress test the broker then you'd want to > understand how to manage the configuration to meet the demands of the > testing. > > -- > Tim Bish > Sr Software Engineer | RedHat Inc. > tim.b...@redhat.com | www.redhat.com > twitter: @tabish121 > blog: http://timbish.blogspot.com/ >
