Stefan Lucke schrieb: > On Samstag, 2. Juli 2005 13:21, Luca Olivetti wrote: > >>Martin Wache wrote: >> >> >>>I thought about letting the softdevice sleep even when the buffers are >>>not full, but I think the correct solution would be that vdr sleep in >>>case it doesn't have some frames to send. Or did I get the Poll() >>>function wrong? Should Poll() sleep in any case? >> >>from device.h >> >> virtual bool Poll(cPoller &Poller, int TimeoutMs = 0); >> ///< Returns true if the device itself or any of the file handles in >> ///< Poller is ready for further action. >> ///< If TimeoutMs is not zero, the device will wait up to the >>given number >> ///< of milleseconds before returning in case there is no immediate >> ///< need for data. > > > What about this interpretation: > > So the question is when has the device an immediate need for data ? > This is the case when the buffer of the device is empty. If this > condition is met, the device has to ignore a given timeout and > return immediate. If buffer empty condition is not satisfied, > device has to sleep when timout is not zero. > I don't think that this the correct interpretation. When the buffers are empty, we have a buffer underrun. We'll never get a smooth playback if we let the buffers run empty. If we define the immediate need for data different (maybe 1/3 of the buffer filled) we will effectivly make the buffers smaller, since as soon as we sleep after/before the reception of one packet we will reduce the throughput to a level far below the consumption rate. The smallest sensible sleep is about 3ms that means we will be able to transfer only 13 packets per frame (40ms). Usually a frame has more packets than 13, so the buffer fill will soon drop below our immediate need for data limit. No, the longer I think about it the more confident I get that the only sensible solution is that vdr should sleep in case it doesn't have any data. Martin
