Hello list, I am new to this list, so please forgive me if I am barging in. I have been using ekiga 2.0.1 for a couple of weeks now, and will continue to do so because I think this is an excellent piece of software, Well Done! On my system I have been experiencing gaps and clicks in the audio while in a call. This tends to occur while graphics operations are happening, eg when switching between windows, or dragging windows around. After much searching on the net, and on mailing lists, I found people describing similar problems, and it seems that 2.6 kernels, certain graphics drivers (eg radeon), and low latency audio don't mix too well. I have also tried all the usual stuff like changing PCI latencies, interrupt priorities, and changing the real-time scheduling attributes of the ekiga process. After a lot of testing, I came to the conclusion that the dropouts in the audio were occurring because the ekiga process, under slight load conditions, was not delivering the audio on-time, and not reading the audio on-time. Looking more closely at what was going on in pwlib, I could see the buffer time was something quite small like 40mS. On my system, in order to eliminate the dropouts on audio playback, this had to be increased to 300mS. This solved the problems on audio receiving audio, but the audio from my mic was still choppy with slight CPU load. Unfortunately, it wasn't just a case of increasing the buffer size for recording because for some reason alsa would only allow 2 periods in the record buffer. Therefore a separate buffer thread was required in order to fix the recording problems. The buffer thread uses high priority realtime scheduling (SCHED_FIFO), and seems to have been quite successful in completely solving the problem with choppy recording. At the moment, I am testing this mod as a proof-of-concept fix, and all seems well so far. I appreciate that the problems I have been encountering are most likely related to hardware/drivers, and the mods I have been doing are all within pwlib source, and not ekiga or opal. If anyone is interested, let me know and i'll post the code/patch, but I must warn you, this is a real hack, the code is very very bodgey. Regards, -- Craig Shelley EMail: craig microtron org uk Jabber: shell jabber earth li
Attachment:
signature.asc
Description: This is a digitally signed message part