Re: a joint letter on low latency and Linux

[Larry McVoy]

> >Paul, what is your definition of "hard realtime"?
> 
> I don't have one. I don't think its helpful to try to use categories,
> when its the actual numbers that matter. I have absolutely no doubt
> that stock Linux 2.whatever can guarantee a response to some external
> event within 10 seconds, short of a system crash or hardware
> failure. 

Actually, that's not true.  Allocate a buffer that is > main memory size
and then sit in a loop an touch it over and over.  You'll see longer than
10 sec delays.

> For me, the interesting question is not "is audio a hard realtime app
> or not?", its "if the response time is (say) 3ms, can
> <some-version-of-linux> guarantee it 99% of the time ?" (100% would be
> better, but 99% will almost certainly do).

That might be OK for home use but is a joke for professional use.  Can you
imagine Digidesign saying "our sound cards work 99% of the time and that
is good enough"?

> if you're going to define *any* guaranteed response as "hard
> realtime", be my guest. you might even be right.

That would be the commonly accepted definition in the literature and in
practice for the last 30 years or so.  Actually, it's closer to "any
requirement for a response that must happen in any fixed interval of
time".  Hard real time has nothing to do with how long the interval is,
it has everything to do with not missing the deadline.