Montavista's "Open Source Real-Time Linux Project" again

Karim Yaghmour karim at
Thu Oct 14 19:13:01 CEST 2004

Paolo Mantegazza wrote:
>     "This project builds on MontaVista's record ... blah ... blah 
> ..................
>     The goal of this project is to deliver performance and deterministic 
> real-time responsiveness that is comparable to that of other commercial 
> and proprietary RTOSes. MontaVista expects to lower Linux task 
> preemption latency (worst case) from a range of one to tens of 
> milliseconds, down to tens of microseconds, for time-critical 
> applications. This would be an improvement of at least two orders of 
> magnitude in Linux responsiveness."

This line of PR is based on marketing slides which show graphs of
"maximum interrupt disable". IOW, they stress a system and then decide
that it's hard-rt based on a sample of data they've obtained by
measuring the system's maximum disabling of interrupts. So, for sure
they can pump-out press releases stating that they have tens of
microsecond response times. The problem is their basic logic is flawed:
to qualify as hard-rt, it must be demonstrated mathematically/
algorithmically that a system is truly deterministic. In order to meet
such criteria, the proponents of these preemption-on-steriod patches
would need to show that _ALL_ software written for Linux in the past
is now somehow compatible with hard-rt.

I may be totally wrong, but I suspect that you can't magically turn a
body of code into hard-rt unless it was architectured with time and
resource constraints in mind to start with. No amount of macro
redifinition or interrupt threading changes the fact that Linux was not
built with hard-rt in mind.

That's without even discussing the fact that the already included very
basic preemption patch is only marginally used, for a variety of reasons,
one of which being stability ...

Author, Speaker, Developer, Consultant
Pushing Embedded and Real-Time Linux Systems Beyond the Limits || karim at || 1-866-677-4546

More information about the Rtai mailing list