19 Jun
2014
19 Jun
'14
9:17 p.m.
hi leonardo,
Next week I'll be in Pisa, Italy, for a workshop
held by some of the
SCHED_DEADLINE guys. I'm not a serious dev but I do research and I'll be
glad to evaluate the benefit of SCHED_DEADLINE for audio and jack,
compared to SCHED_FIFO.
that would be nice! the interface is much more robust than SCHED_FIFO
and it is rather similar to time-constraint threads as found on mach
threads. and it perfectly maps to rate-monotonic schedulers as found in
audio systems ...
however my initial tests were not very promising: my multicore audio
engine performed worse when using SCHED_DEADLINE for the helper threads
compared to SCHED_FIFO ...
cheers,
tim
> since recent kernels provide SCHED_DEADLINE,
i'm porting my code to make
> use of it.
>
> * is there any plan to migrate jack1 and/or jack2 to use this scheduling
> policy for the jack thread(s) of the application?
>
> * what is the recommended way to set the scheduling policy of the jack
> thread with the current API? afaict, the JackThreadInitCallback is
> called before JackClient::AcquireSelfRealTime in jack2. (i cannot use
> jack1 due to the old bug that jack corrupts the stack while trying to
> pre-fault it)