From: brian g. <bg...@us...> - 2003-01-14 21:28:32
|
On Tue, 14 Jan 2003, Richard Vaughan wrote: > > The method of checking timestamps works well for periodic data, but > doesn't seem quite right for random data like comms. The broadcast comms > packet isn't all that big; why not store a checksum to test for newness > instead of a timestamp? (Include the timestamp in the checksum to handle > repeating data packets). A little overhead, but it'd make problems like > this go away. > Not a bad idea, but one that would require some changes to the server. We would have to add a checksum hook into the driver API, by which the server could check for newness. Alternatively, we could fake it by having comms drivers stuff their unique checksum into the timestamp slot, likely at the expense of losing the timestamp information on the packets. I'm not necessarily oppposed, but I am hesitant about adding more stuff to the driver API. Besides, this problem only arises with comms drivers in Stage (or if you manage to receive messages at a rate of more than 1/microsecond, which seems unlikely). Now, that 'only' might describe a great many users, in which case there should be a solution. I suspect that the PUSH_ALL mode should suffice, but then i don't actually use the comms drivers. brian. |