Potential message size performance issue
----------------------------------------
Key: AMQ-1387
URL: https://issues.apache.org/activemq/browse/AMQ-1387
Project: ActiveMQ
Issue Type: Bug
Components: JMS client
Affects Versions: 4.1.1
Environment: RHEL4 Update 4 x86_64, Java 1.6.0_02, 3.6 GHz Xeon, 4GB
RAM, 2 Dual core ( 4 core total )
Reporter: Bob Tillson
I have a JNI client that interfaces to JMS.
Tested previously on JBOSS and JORAM; functionally correct and performs
adequately.
"Porting" to activeMQ. No coding changes, just use different jar file as all
through generic interfaces.
So far so good...
Start testing, have publisher and subscriber apps for testing.
publisher can vary message size, so start at 16 bytes, and double message size
up to 1MB for basic functional verification using ActiveMQ 4.1.1.
publisher transmist message "as fast as possible", no sleeps.
Everything seems reasonable up til about 4KB messages; at 4KB, get ~ 4200
msgs/sec, seems reasonable vs other paltforms; using large amounts of CPU, but
that is OK.
8K message size, only get 25 msgs per second throughput. Seems to be a magic
number.
Above 8K, message throughput drops, but also CPU usage drops considerably on
publisher, subscriber, and server. at 64K, less that 80 msgs per sec
throughput...
Server cpu dropping not suprising cause many less messages being pushed through.
The real issue is why is the publisher CPU dropping. It is consistently below
10 percent.
At 128K message size, speed is a bit faster close to 100 msgs per sec.
Seems like there is some REALLY werid stuff going on in the client side...
Has anyone lese run into this?
Am I just freakin crazy? Especially the 8K message size and then the bigger
the message, the faster is gets...
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
