On 11/06/13 07:16, "Thomas B. R?cker" wrote:> Hi,>> Other datail: all users have always "Lag: 0" on Admin interface > > That doesn't need to be a problem, but without full understanding it's > hard to say.that is just the difference between the latest source offset and client offset, at 0, just like on the main trunk, the listener is at the latest point on the queue. Simply means that there is no lag on the icecast to listener link> I've had this stream now running for about 1800s and have observed one > peculiar thing, the cache-fill went constantly down from an initial 20% > to now about 3%. > I let it run a bit longer and this is what I get: > > Cache empty, {?} > A:1938.8 (32:18.8) of 0.0 (unknown) 1.4% 0% > > I suspect that Winamp is signalling a low buffer by that red square. > So, yes, that's why the players drop, the bitrate is a tiny bit lower > than the playback rate. This means that over time a player will deplete > the cache and inevitably either rebuffer or start to stutter. You can > also force this to happen immediately if you disable buffering completely. > As to the root cause I can only speculate it could be some KH specific > feature or it can be that your source client has a problem.saw exactly the same thing yesterday and the stats for the incoming bitrate indicate a slightly lower than expected bitrate Just for completeness, the sam is apparently transcoding from files, not live input and is also within virtualbox, the icecast is on not within a VM. While I don't know for sure, I suspect sam is probably using the VB clock for regulating the stream and if there is some issue on drift then that would account for it. karl.
Hi, On 11 June 2013 12:01, Karl Heyes <karl at xiph.org> wrote:> On 11/06/13 07:16, "Thomas B. R?cker" wrote: >> I've had this stream now running for about 1800s and have observed one >> peculiar thing, the cache-fill went constantly down from an initial 20% >> to now about 3%. >> I let it run a bit longer and this is what I get: >> >> Cache empty, {?} >> A:1938.8 (32:18.8) of 0.0 (unknown) 1.4% 0% >> >> I suspect that Winamp is signalling a low buffer by that red square. >> So, yes, that's why the players drop, the bitrate is a tiny bit lower >> than the playback rate. This means that over time a player will deplete >> the cache and inevitably either rebuffer or start to stutter. You can >> also force this to happen immediately if you disable buffering completely. >> As to the root cause I can only speculate it could be some KH specific >> feature or it can be that your source client has a problem. > > saw exactly the same thing yesterday and the stats for the incoming > bitrate indicate a slightly lower than expected bitrate > > Just for completeness, the sam is apparently transcoding from files, not > live input and is also within virtualbox, the icecast is on not within a > VM. While I don't know for sure, I suspect sam is probably using the VB > clock for regulating the stream and if there is some issue on drift then > that would account for it.That would explain it. VMs are notoriously bad choice for running source clients due to timing jitter. We've seen malformed Ogg streams and other oddities in the past. So having a slightly slow clock is no surprise to me. Cheers Thomas
On 11/06/13 17:18, Thomas R?cker wrote:> Hi, > > On 11 June 2013 12:01, Karl Heyes <karl at xiph.org> wrote:>> Just for completeness, the sam is apparently transcoding from files, not >> live input and is also within virtualbox, the icecast is on not within a >> VM. While I don't know for sure, I suspect sam is probably using the VB >> clock for regulating the stream and if there is some issue on drift then >> that would account for it. > > That would explain it. VMs are notoriously bad choice for running > source clients due to timing jitter. We've seen malformed Ogg streams > and other oddities in the past. So having a slightly slow clock is no > surprise to me.I have had feedback. The VirtualBox (latest) has some issue on the timing side of things. Using Hyper-V has not had the same problem. Previous VB versions has not had this issue either so something has broke there karl.