On 2009-12-17 at 14:32:21 [+0100], Axel Dörfler <axeld@xxxxxxxxxxxxxxxx> wrote: > Ingo Weinhold <ingo_weinhold@xxxxxx> wrote: > > > revision yet. The recent libmedia and mp3 reader changes might > > > already have > > > improved the situation. > > Kind of a mixed bag... The changes generally seemed to have improved > > lateness of buffers arriving at the mixer. Even with a -j8 build > > running > > there are rarely any buffers arriving with greater lateness than 1 ms > > (which > > I use as the output threshold). At the beginning of an audio track > > there are > > always several buffers late (usually between 10 and 20 ms). > > Unfortunately > > the occasional audio skips (which also still happen in an idle system) > > are > > no longer only a few 0.1 s long, but extend over several seconds > > (sometimes > > as worse as a minute!). :-/ > > I wouldn't have expected such an impact (on either end). A greater > chunk size should not be as CPU intensive, but that should be about all > it does (although this also depends how the media player caches its > chunks). > Have you tried reverting to the old ChunkCache implementation with the > changes in the mp3 reader? Does this already cause the large hickups? After replacing my debug_printf()s by syslog()s things are back to normal (i.e. short skips). Apparently the debug_printf()s were eating so much CPU that they dragged the catching up on quite a bit. CU, Ingo