|Microsoft DirectX 9.0|
Messages often vary widely in importance. Some are time-critical and must be delivered as quickly as possible. Others can be delayed if necessary or, possibly, not sent at all. One issue with congestion control algorithms is that an application might create messages faster than they can be sent. Unsent messages must then be held in a queue until an outgoing slot opens up. If all unsent messages are held in a single pending-message queue, high priority messages might be blocked while waiting for lower priority messages to be sent first.
The Microsoft?DirectPlay?protocol solves this problem by having three pending message queues: low, medium, and high priority. When a slot opens up in the sent-message queue, the protocol selects the next message to be sent as follows:
This priority mechanism enables you to get your time-critical messages out as quickly as possible, even when less important messages have already been submitted.
One of the consequences of throttling is that messages might spend a relatively long time in a pending-message queue, especially if they are low priority. Some messages might stay long enough to have been superseded by subsequent messages. These messages are no longer relevant. For example, your application might periodically send player-location update messages. Each update is independent of the others and supersedes any previous updates. If you have two player-location updates in the pending message queue, only the most recent one needs to be sent.
The DirectPlay protocol enables you to handle the problem of outdated messages by adding an optional timeout value to the message. If the message is still in a pending-message queue when the timeout expires, the message will be canceled.
When an application sends a disconnect message, the message is placed at the end of the low-priority pending-message queue, and the protocol stops accepting outgoing messages. This practice guarantees that all pending messages are sent before the link is disconnected. The disconnect message is sent as a reliable sequential message to guarantee that it arrives, but not before all other messages in the queue have been delivered.