Re: RTCore and RTLinux

[tim bisson <tbonecoder@xxxxxxxxxxxx>]

Hi,

I mostly agree with you.

Mike Ji wrote:

> Hi, folks,
>
> I happen to read some white papers related to RTLinux. Below is some of my
> personal thoughts. 
>
> * RTLinux's so called realtime extension-RTCORE  **was* to address the
> previous issue--Linux is not realtime supported, for example, fully
> kernel-preemptive; posix realtime extension.
>
> * After linux 2.6, I did not see any outstanding reason that people still
> need to use a RTLinux **patch** to achieve realtime applications.
>  
There are still several reasons 2.6 is not suitable for real-time 
applications:
-scheduler latency is non deterministic ( time between interrupt firing 
and process execution waiting
on that interrupt). The rtirq patch helps by adding priorities to 
interrupts.
-Linux scheduler isn't real-time - only such schedulers such as EDF/RM 
will give you that.
-Data acquisition from disk is also non deterministic. This issue is 
much harder to deal with because its
almost impossible to bound the amount of time it takes read/write data, 
since you can't tell where the disk
arm is relative to how the data is placed on disk. Most real-time apps 
get around this simply by not writing
to disks.
-Caching hurts real-time application performance, in that running two 
real-time apps might cause the cache
to get continuously flushed (because the same cache lines are used), 
while running only one real-time app
might cause zero flushes. A solution is page coloring, but Linux 2.6 
doesn't have this. There are patches, though.

> * From os/kernel viewpoint, it is very hard for me to position RTCORE, but a
> realtime patch. I mean, RTCORE is not able to be identified into any
> well-known os architecture. 
>  
Sorry, I don't really know what you mean.

> * If there is a very high performance micro-kernel, say, L4 or QNX/Neutrino,
> I personally think it is much better than the RTCORE stuff, which HAVE TO
> bind a secondary os as running time environment.

I agree the hierarchical approach isn't the correct model for a 
real-time system.  Making Linux is deterministic is the
right solution, but implementing that is hard.

L4 is the Real-time operating system from the Research group in Germany, 
right? Doesn't it too require the use of a
micro kernel, such as fiasco or pistachio.

Has anyone played with the nucleus kernel from accelerated technology? 
If so, how did they like it?

More personal thoughts,
Tim

> Some personal thoughts,
>
> --
> Kernelnewbies: Help each other learn about the Linux kernel.
> Archive:       http://mail.nl.linux.org/kernelnewbies/
> FAQ:           http://kernelnewbies.org/faq/
>
>
>  


--
Kernelnewbies: Help each other learn about the Linux kernel.
Archive:       http://mail.nl.linux.org/kernelnewbies/
FAQ:           http://kernelnewbies.org/faq/