Re: a joint letter on low latency and Linux

[Cort Dougan]

I was referring to the case of:

  write(fd, buffer);
  do_some_heavy_fp();

do_some_heavy_fp() might have to wait a long time to run, since it follows
some disk IO.  This is the sort of thing we couldn't allow in any
hybrid-ized Linux/RTLinux API and isn't the sort of thing your time
critical code is likely to be doing, anyway.

} Doing various kinds of heavy disk IO shouldn't effect how the machine
} does math computations. It's silly otherwise.

I'm interested more in what we could offer in another API that looks more
like Linux to make moving large projects like this over.  We can give you
hard latency guarantees already.  Far more than you need - but that doesn't
hurt you it just allows you to do more of your processing before your
deadline hits.

We have what we believe to be a very good hard realtime programming model.
It forces you to decouple your realtime code and your non-realtime code.
What we'd like to do is allow RTLinux to use the structure your normal
Linux code already has to do this implicitly by executing your code in
RTLinux but not forcing you to adopt the RTLinux programming model.

} I'll take a look at it what you've guy have done recently with the API. :-)