Hi Emmanuel,
I did some tests using the PingDelayInM, but without noticeable improvements.

With respect to the Nagle algorith, using Linux, I noticed no visible 
performance differences (CPU ~ 0%).

Instead, I could get a 0% CPU load (yes, I've written it correctly!) with a 
RHEL 4 host running natively, and it sounds little strange for me (and maybe 
normal for you ^__^ ). The same operating system, running on a VirtualBox 
virtual machine, gives a 0% load, but the VM's instance uses 100% of CPU. May 
this is a "bug" of RHEL top util? Or some RedHat specific kernel patch? Or 
some incompatibility/regression/issue with new JVM and new kernels?

Danilo

Il Tuesday 29 April 2008 17:44:30 Emmanuel Cecchet ha scritto:
> Hi Danilo,
>
> The driver is supposed to ping only every second according to the
> default settings but you can try to override the value by appending the
> following option to your JDBC URL: ?pingDelayInMs=60000 to force the
> ping to happen only every minute.
> Unfortunately, it looks like there is no way to turn this pinger thread
> off. Also it looks like this fires daemon threads which can cause
> problems if the JVM never stops (it will keep forking daemon thread each
> time a new URL is accessed).
> Another thing worth mentioning is that Sequoia disables the Nagle
> algorithm on TCP connections (socket.setTcpNoDelay(true)). It would be
> worth trying that on your client/server program to see if that changes
> the performance. This might have a greater impact on Windows than Linux.
>
> > As you suggested, I played a little with networking options. First of all
> > I did some more tests disabling group communication (no <Distribution>
> > element in virtualdatabase config file, am I right?) , without any
> > changes in results: [EMAIL PROTECTED] and receive0 with predominant cost.
>
> Yes, this is right.
>
> > May you suggest me other tests?
>
> Keep me posted with the results you obtain with the 2 options above.
>
> Thanks for your feedback,
> Emmanuel


_______________________________________________
Sequoia mailing list
[email protected]
https://forge.continuent.org/mailman/listinfo/sequoia

Reply via email to