Hi,
 
In continuation with previous email, more information regarding the issue:
we have analysed the logs and observed that there is some pattern of varying
publishing times, for every  few seconds.
 
example:
At the full load, 
the event published at 14:49:24,352 took TotalTime : 55 ms
and then gradually increased..and 
the event published at 14:49:26,445 took TotalTime : 4475 ms
 
Hence within this approx, 2 minutes duration, there were 190 response
messages published to 
ResponseTopic; starting at 55 ms and increasing upto 4475 ms..
and after that,
the next event published at 
14:49:26,498 took only 62 ms;
 
from thereon, again the same pattern is observed...
 
Is there any JMS resources contention happening in this duration between the
publisher and the ActiveMQ standalone broker ? 
as known, send API() uses tcp connection from the application instance to
the standalone broker instance.
 
is there any TCP tuning to be done ?
or
is there any Active MQ tuning to be done ?
 
Both Active MQ JVM heap and the application JVM heap are quite stable. there
are minor collections
happening every few seconds.
 
Note: JMeter test plan is being run for 4000 conc users with think time of 5
seconds; and jmeter throughput arround 450 req/sec  
 
please let us know if anybody has clues ?
 
regards




Y wrote:
> 
> 
> Hi, 
> 
> Thanks for the reply . 
> 
> When observed in Broker Statistics in Jconsole it has enough memory. 
> 
> Here is my ActiveMQ.xml file .. 
> http://www.nabble.com/file/p23539624/activemq.xml activemq.xml 
> 
> Here is the Memory Snapshot of ActiveMQ Old Generation.  
> http://www.nabble.com/file/p23539624/OldGen.JPG OldGen.JPG 
> 
> 
> 
> 
> Gary Tully wrote:
>> 
>> can you give more memory resources to the broker? I imagine there is
>> some spooling to disk under high load. It may make sense to post your
>> activemq.xml.
>> 
>> 2009/5/14 KumarR <rykum...@gmail.com>:
>>>
>>> Hi All ,
>>>
>>> I am having problem with publishing messages to Topic. It's taking too
>>> much
>>> time to publish a message. Below are the details.
>>>
>>> Test Case:
>>>
>>> The Client  is publishing messages to RequestQueue. and waiting
>>> responses on
>>> ResponseTopic using a co-relation id.
>>> Server program consuming messages from RequestQueue using Spring Message
>>> POJOs. These POJOs after consuming and processing the messages,
>>> forms the response message and publishes to the ResponseTopic using
>>> co-relationid.
>>> The client using blocked receive(TIMEOUT) receive responses from
>>> ResponseTopic using Message selector with co-relation id.
>>>
>>>
>>> Observations: at higher loads, around 3000 conc users, the publishing
>>> response messages back to Response Topic becoming slow;
>>> taking around 5 secs to 10 secs. otherwise in 2000 conc users load, it
>>> was
>>> taking around 100 ms-120 ms.
>>>
>>> Setup Configuration:
>>> We are using ActiveMQ 5.2 on Sun Solaris 10 - SPARC JVM 1.6_16;
>>> Topic has Non-Durable Consumer;  we use Non-Transactional,
>>> Non-Persistent
>>> Messages
>>>
>>> Client program is a WebApplication running on JBoss-JVM instance on
>>> server
>>> machine 1
>>> ActiveMQ is a standalone broker jvm running on server machine 2
>>> Server program is another backend application (not web-application) on
>>> another Jboss-JVM instance on server machine2.
>>>
>>>
>>> Can any one help in solving this problem ?
>>>
>>> Thanks
>>> Kumar
>>>
>>> -----
>>> Thanks,
>>> Kumar Ravella
>>> --
>>> View this message in context:
>>> http://www.nabble.com/Publishing-to-Topic-is-very-slow-in-higher-loads-ActiveMQ5.2.-tp23538009p23538009.html
>>> Sent from the ActiveMQ - User mailing list archive at Nabble.com.
>>>
>>>
>> 
>> 
>> 
>> -- 
>> http://blog.garytully.com
>> 
>> Open Source SOA
>> http://FUSESource.com
>> 
>> 
> 
> 


-----
Thanks,
Kumar Ravella
-- 
View this message in context: 
http://www.nabble.com/Publishing-to-Topic-is-very-slow-in-higher-loads-ActiveMQ5.2.-tp23538009p23596007.html
Sent from the ActiveMQ - User mailing list archive at Nabble.com.

Reply via email to