Re: a joint letter on low latency and Linux

[Cort Dougan]

This is something RTLinux is for.  You can get a worst-case upper-bound on
the time your code gets the cpu from the time a given event occurred.  Even
more important, you have it available to you _right now_ with RTLinux.  Of
course, porting your code to use the RTLinux API can't happen _right now_
:)

What exactly makes RTLinux porting difficult for you?  I mean
specifics (we can take this discussion off the list...).  We're looking at
how to make the RTLinux API more acceptable and make migration easier for
general Linux programs (as well as pSOS and VxWorks).  If you can give me
some good examples of what makes RTLinux porting difficult we can move much
more quickly.

At one extreme we've been looking at providing a mechanism similar to what
RTAI gives programmers.  We can make your program look like it's running in
a normal pthreads environment without the RTLinux API as long as you stick
to some constraints such as not doing file IO and other things incompatible
with realtime (things you're not likely to do in time critical code
anyway).

} Exactly. The result of the intensive (read: frequently buggy and hard to
} write code) has to make it back out with low latency (i.e. 99.9% of the
} time <=5ms)