drneau.com

[drneau]

Look for coming improvements in the client/server robustness...

The way I did it for v1 (and copied for v2) worked, but I've come to the conclusion that it needs reworking.

Boring inner-working Java details to follow.


The problem is that all the following happens within one thread:
- Segment timer decrements "remaining time property"
- Property changes and notifies all property change listeners
- The "ConsoleDataModel" catches the change and notifies its listeners
- The server process catches that change and updates each client, one at a time

The problem here being that if a connection to a client hangs, then the server hangs. That, and it's just simply inefficient...

Under the new model, the following will happen in one thread:
- Segment timer decrements "remaining time property"
- Property changes and notifies all property change listeners
- The "ConsoleDataModel" catches the change and notifies its listeners
- The "ServerChangeListener" will catch the change and pop it onto its message stack.

...while the following happens in another thread:
- The "ServerStackWatcher" checks the stack for new messages.
- If a message is found, it is routed to the main server process which in turn pops the message onto each client's message stack.
- Each client watches its own stack. When a message is found, it gets routed.

This way, clients only hang themselves.
_________________
--
Dr. Neau (not a real doctor)
http://drneau.com
http://home.comcast.net/~jneau

[drneau]

Implemented the new mechanism in the 4/7/2005 release.
_________________
--
Dr. Neau (not a real doctor)
http://drneau.com
http://home.comcast.net/~jneau