Another cpu and memory consuming hard to find

[Frederic Bregier]

Hi Trustin,

I've downloaded the last JDK6.
Now the results are a bit different:

-JDK5 : (200 Clients every 100ms) 33 MB/s of bandwith, 730x3 Msg/s in both
way,
  one way (client->Server) using msg of approximatively 150B, other way 
  (Server->client) using msg of approximatively 16KB, 74 ms from jmeter
point of view
  (for 3 msg exchange), and 74 MB of memory in the process

  Note that in the log I've got:
2008-12-01 10:12:44,138 INFO [main]
org.jboss.netty.channel.socket.nio.NioProviderMetadata [LoggingHandler.java
info : 120] Using the autodetected NIO constraint level: 1 (Use better NIO
provider for better performance)
2008-12-01 10:13:35,902 DEBUG [New I/O server boss #1 (channelId:
f1d233b9-011d-1000-b8f8-01a494bbfe33, /0:0:0:0:0:0:0:0:6666)]
org.jboss.netty.handler.execution.MemoryAwareThreadPoolExecutor
[LoggingHandler.java debug : 104]
ThreadPoolExecutor.allowCoreThreadTimeOut() is not supported in this
platform.


-JDK6 : (200 Clients every 100ms) from 8 MB/s to 0.6 MB after 12 000
requests of 
  bandwith, from 150x3 to 12x3 Msg/s in both way,
  from 1500 to 15 000 ms from jmeter point of view (for 3 msg exchange),
  and immediately 129 MB of memory in the process

  In the log I've got:
2008-12-01 09:53:08,937 INFO [main]
org.jboss.netty.channel.socket.nio.NioProviderMetadata [LoggingHandler.java
info : 120] Using the autodetected NIO constraint level: 2 (Use better NIO
provider for better performance)

  I've got no more 100% of CPU on one CPU. When I stopped the jmeter, the
CPU goes to 
  almost 0 (0.7% but not 0, if I ask the client to close its connections, it
goes to 0),
  the memory does not increase, I can stop the process as normal.

I've made two dumps (full this time), composed by several files.
One dump was done before closing all connections (cpu at 0.7%), and one
after
(cpu at 0%) on the same process. They are quite big...
So I remove the core part of the dumps.

before  http://n2.nabble.com/file/n1598774/dump-before-close-conns.tar.gz
dump-before-close-conns.tar.gz 
after  http://n2.nabble.com/file/n1598774/dump-after-close-conns.tar.gz
dump-after-close-conns.tar.gz 

I didn't try right now to open those files...

I try to make a simple example but it didn't reproduce the behaviour
until now (70MB sustained).

HTH,
Frederic



-----
Hardware/Software Architect
-- 
View this message in context: http://n2.nabble.com/Another-cpu-and-memory-consuming-hard-to-find-tp1558054p1598774.html
Sent from the Netty Developer Group mailing list archive at Nabble.com.