Another cpu and memory consuming hard to find

Trustin Lee tlee at redhat.com
Tue Dec 2 20:42:19 EST 2008


As a last suggestion, you could try to increase
NioSocketChannelConfig.writeBufferHighWaterMark and
writeBufferLowWaterMark property?  I guess it won't make a lot of
difference but it worths a shot.

Thanks for keep trying to analyse the problem.  I really appreciate your
effort!

Cheers,
Trustin

On Wed, Dec 03, 2008 at 01:09:08AM +0900, Frederic Bregier wrote:
> 
> Hi Trustin,
> 
> I think we should close this thread... ;-(
> 
> I try several things:
> - I implement a client that makes 100 connections (threads) in parallel and
> act
>  (outside Tomcat, so directly in its own jvm) as a test of benchmark.
> - Still JDK5 perfect : almost 500 Mbit/s
> - Still JDK6 AIX IBM defect : from 80 Mbit to 5 Mbit after a thousands of
> requests
> - I download the last version for Linux of both Sun's JDK and IBM's JDK in
> version 6.
>   - JDK5 IBM : ok (smaller server so 80 Mbits constant)
>   - JDK6 Sun : ok (same)
>   - JDK6 IBM : ko (from 16 Mbit to 5 Mbit and decreasing)
> 
> I have no more memory growing or cpu consuming.
> 
> - I try to make a simple example, but it does not reproduce the behaviour...
> ;-(
> 
> So I have no more clue.
> I think the JDK6 from IBM (both for AIX and Linux) seems buggy...
> 
> I will continue to use JDK5 (on AIX I have no choice to use another JDK).
> And I will wait for new patches from IBM on its JDK6.
> I don't feel like I can make a call to the support of IBM since it is not
> easy to see where is the problem (and it seems really in the JDK).
> 
> Except if you have any other ideas, I will stop this thread for now...
> 
> Frederic
> 
> -----
> Hardware/Software Architect
 
-- 
Trustin Lee, Principal Software Engineer, JBoss, a division of Red Hat
--
what we call human nature in actuality is human habit
--
http://gleamynode.net/
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 197 bytes
Desc: not available
Url : http://lists.jboss.org/pipermail/netty-dev/attachments/20081203/5bec5cbc/attachment.bin 


More information about the netty-dev mailing list