Re: [linux-audio-dev] Re: costs of IPC

[Paul Davis]

>> it means that you can get the same safety with a single-thread
>> approach.
>
>In multi thread you can kill/restart a blocked component without to
>restart the whole engine.
>
>With single thread this is not feasible

disagree. assuming you can detect the blocked component, you just
avoid running it any more (i.e. remove it from the list of active
components). you might also instantiate a new instance of it (though
this is unlikely to solve whatever problem caused it to block).

we do something like this in ardour, though its user driven - i have
to frequently "turn off" steve's retroflanger because its not
RT-friendly with certain settings. you can turn it off, adjust the
parameters, turn it back on. it works fine. it doesn't handle
"blocking", but we've already established that blocking would need to
be handled by an asynch timer.

--p