[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [xmlblaster-devel] Added updateOneway() and publishOneway() support

You did a good job.

On Mon, 18 Mar 2002, Marcel Ruff wrote:

>Hi again,
>there is now support to publish() and update() messages
>in oneway mode.
>Oneway invocations don't have a return value and
>can't throw an exception over the socket.
>Normal method invocations have a return value
>and throw exceptions. This is kind of an
>application level ACK/NACK.
>The 'oneway' variant relies on the TCP ACK,
>which is reliable as well, but does not allow
>transaction safety.
>Note that CORBA and our SOCKET protocol
>support native 'oneway', whereas we simulate
>oneway with RMI and XmlRpc (not much
>performance gain).
>The EMAIL driver is always oneway as well
>as is Heinrichs letter dove :-)

Actually this was more an idea about an implementation of a physical
layer than a communication protocol ;-)

>Every publisher and subscriber can choose which
>variant it wishes to communitcate.
>The performance boost is promising, on my
>600 MHz Linux box the oneway publishing
>(with burst mode) reached far over
>    3000 messages/second.
>using CORBA.
>This is fun.
>The next step is to tune the raw SOCKET connection
>to beat CORBA.
>The future xmlBlaster clustering framework will
>possibly use the SOCKET protocol (it is configurable)
>to allow tunneling firewalls and probably proxies
>with real time callbacks.
>Yes, this is fun :-)