Subject: Re: [linux-audio-dev] Realtime restrictions when runningmultipleaudioapps .. Was: Re: disksampler ...
From: Paul Barton-Davis (pbd_AT_Op.Net)
Date: Wed Jul 19 2000 - 21:42:49 EEST
> it looks like the overall idea would be to have sort of two-level
>user-space sound server - low latency part would trade some stability
>for performance (the applications would be loaded into one process
>space), the 'general' part would offer better stability (masking the
>crashes of low latency part).
> seems to make some sense. does it?
i think so, with the proviso that the "low latency part" is not really
client/server in any conventional sense, but is really just dynamic
linking and unlinking of code. the only part that would be at all like
a client/server design is an API for GUI's etc to do their thing.
> what latencies do we hope to get for both levels? <5ms (almost
>guaranteed) for low latency level, <15ms (with slightly weaker
>guarantee) for general level?
You won't be able to put accurate bounds on the general level, since
each time you context switch, the kernel may decide to run something
else instead. If you have a time when a lot of applications are ready
to communicate with the sound server, it could be an arbitrary time
until they all finish doing so. But I would have thought that 20-40ms
would be pretty easy.
--p
This archive was generated by hypermail 2b28 : Wed Jul 19 2000 - 22:23:14 EEST