Oh and reliable delivery is something that should be optional IMO.

Mark.

On 19 Jun 2007, at 00:25, Ron Sigal wrote:

Hi Tim,

Excellent. This is just the kind of feedback I was hoping for. We definitely want Remoting 3.0 to satisfy the needs of JBM. You're our best critics. :-)

-Ron

Tim Fox wrote:
Ron Sigal wrote:

Scott M Stark wrote:

The main problem for me with the TowardsGreaterSymmetryInRemoting page
is that its not talking about a base asynch message oriented
architecture. Much of the current asymmetry's is due to the rpc oriented
api. If you flip this around to have a base asynch message view, all
communication is handling of these messages. RPC with callbacks is
setting up blocking message handlers. Symmetry from a higher level
Client api is also not a requirement in my view. By definition a
callback is an unpredictable event/out of band msg with respect to some
rpc call returning a value. The use of client and server are also by
definition asymmetric and map to msg senders/receivers. We need to start
from the bottom and move back up to the rpc api in order to be able to
talk about what the 3.0 version of Client should look like.

Actually, a "base asynch message oriented architecture" was just what I was trying to get at. While Remoting should continue to support the rpc model, the Connection.receive() and Connection.send() methods that I mentioned are intended to support asynchronous message sending and receiving.

Does Connection.receive() block until it receives a message()?

Remoting 3.0 needs to support non-blocking semantics too to cope with very large numbers of connection (We can't have a thread per connection blocking on receive()).

What you probably need is some kind of select() functionality (see the Java NIO API or unix select() and poll()) where you can register for events - in this case a single (or small group of) thread(s) would register for events on multiple "channels" and are woken up when an evens matches the selector.

You probably also want to build in support for aynchronous IO via callbacks - in this case, you don't even have thread(s) waiting on select() but register some kind of callback handler and the OS calls your handler directly - this can occur with less context switching than select().

Also, while it's true that client and server roles are inherently asymmetric, actors can play multiple roles (like Peter Sellers). In Remoting, for example, callbacks (in push mode) are handled by clients on the server side talking to servers on the client side. I think the same thing would be conceptually simpler with a "connection" abstraction that mirrors a real TCP connection: it's true that there are client and server sockets, but once the connection has been created, there can be senders and receivers on both sides.

The architecture also needs to be layered such that you can plug into
low level message creation for the case of needing to control the on the
wire format of these messages.

We are brining on the MINA lead, Trustin Lee, so we will need to look at
how

The idea of stacks of marshallers and unmarshallers in Remoting has been floating around for a while, and Tom did some initial work in that direction. I'm thinking that's where the layered message handling will live. I've been meaning to write a second document on the subject, but, in fact, MINA has a pretty flexible and sophisticated framework for chains of message handlers, which looks like a good match for what we want. As you say, we need to understand how MINA and Remoting will work together.

Anil Saldhana wrote:

Ron,
Most of it may already be present.

Here is what I am thinking:
a) Pluggable mechanism to do authentication at either ends of the pipes
(SASL)
b) Pluggable ways to secure the payload that passes through the pipes.

Regards,
Anil

Ron Sigal wrote:

There have been various attempts to get some discussion going about
the features desired for the next generation of Remoting, and so far I
think the buzz has broken the -80 db level. I'm trying again with the
wiki page at
http://wiki.jboss.org/wiki/Wiki.jsp?page=TowardsGreaterSymmetryInRemoting. We in the Remoting group (i.e., me in the Remoting group) would like
to hear from the Remoting stakeholders about what features would make
Remoting more usable for you. Of course, I could just go ahead and
write fun stuff. :-)

-Ron

_______________________________________________
jboss-development mailing list
jboss-development@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/jboss-development


_______________________________________________
jboss-development mailing list
jboss-development@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/jboss-development

------------------------------------------------------------------------

_______________________________________________
jboss-development mailing list
jboss-development@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/jboss-development

_______________________________________________
jboss-development mailing list
jboss-development@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/jboss-development

--
JBoss, a Division of Red Hat
"My company's smarter than your company."

_______________________________________________

jboss-development mailing list

jboss-development@lists.jboss.org

https://lists.jboss.org/mailman/listinfo/jboss-development

----

Mark Little

mlittle@redhat.com

JBoss, a Division of Red Hat

Registered Address: Red Hat UK Ltd, Amberley Place, 107-111 Peascod Street, Windsor, Berkshire,

SI4 1TE, United Kingdom.

Registered in UK and Wales under Company Registration No. 3798903

Directors: Michael Cunningham (USA), Charlie Peters (USA) and David Owens (Ireland)