From: Dan M. <dm...@dc...> - 2000-09-20 00:27:52
|
> An interesting idea, but possibly one with only limited application > here. It really depends on the data stream, but if every vertex in an > animation frame differs from the previous one (and that's not hard to > do) you're pretty much stuck. IMHO the right way to do remote rendering is to use some sort of high-level scene graph protocol between client and server. Look at how Quake handles slow internet connections -- the network server transmits very high-level, low bandwidth commands ("Player 5 moves to 123.0,482.1"), which the local client translates into high-bandwidth rendering commands. You could argue that Quake can render at high framerates across a 56k modem line! It's perfectly feasible to send scene graph updates through a socket; you just can't expect to shove every single vertex through the same pipe. (look at where graphics API's are going -- I betcha the next generation chips will be storing large chunks of high-level scene data in video RAM to minimize bus traffic...) Applied to your particular concern: how about a local direct-rendering client that uses one of the many OpenGL scene graph libraries? Make RPC stubs for the entry points and use that as your network protocol =). Dan |