Have I adaquatly described BULLET?

[Jeff]

There are limits on the window size and TCP requires Acks.  That puts a limit on the throughput that becomes somewhat independant of bandwidth, for high bandwidths, and primarily dependant on latency. 

Fair enough.

In practice at TEN we never found this to be an issue.  Ofcourse the hjghest bandwidth we had to deal with for most of our customers was dual-ISDN.  The only real problem we had with TCP were those re-transmit stalls, which gets us abck to the original topic of the conversation

[phi6]

Ok thanks for your replies.

Now I am confused.

Why implement a reliable UDP protocol (at user level) if the performance is worse than UDP when the message in undelivered?

I thought UDP and TCP under normal conditions (when messages are delivered successfully) perform equally, so how does a UDP scheme that is reliable going to help?

[swpalmer]

I thought UDP and TCP under normal conditions (when messages are delivered successfully) perform equally, so how does a UDP scheme that is reliable going to help?

For the throughput required of most games that is probably true.   The UDP schemes that I was talking about are all about getting the highest throughput in a situation of high bandwidth and high latency.  Missed packets in those schemes don't have to be redelivered in a timely manner so long as the total throughput remains high.  So such schemes are mainly useful for bulk transfers, not the realtime needs of games.  (Except maybe for broadcasting game levels and patches where they would actually do very well.)