OS Overhead

[Carl Reynolds]

Burke, Thomas G. wrote:

>All,
>
>	This is a somewhat convoluted question, but here goes...
>
>	We are creating a system that has a "timeline" (e.g. tasks must
>be completed in a given amount of time)
>
>	OK, so we have a processor (we'll stay generic at this point)
>that will be running a linux based kernel, and we want to figure out how
>much processor availability we have (e.g. clock tics in a given time).
>
>	We know that the kernel functions (including resource
>allocation, interrupt hadling, and so on) take up some omount of CPU
>horsepower (albeit small)
>
>	What is a good general rule of thumb for a barebones linux
>system?  I know this is gross estimation, but does the kernel use up 20%
>of my CPU resources?  10%?  Any ideas?
>
>	Thanks
>		Tom
>  
>
Hi Tom,

I have used both QNX and Lynx to develop real time systems. We didn't 
look at the use of resources in the same way you are so I'm not sure if 
I can answer your question. However, we were running systems that had 
latency times on the order of three to five milliseconds. Both QNX and 
Lynx were very capable of handling responses in those time frames. I 
know that these can be considered rather long in some real-time 
applications. Since you have not said what kind of latency you are 
expecting in your application, it is hard to figure out what the demands 
of your product will be.

I have also worked on systems using Linux on an ARM where we were 
experiencing latencies on the order of 100 to 150 milliseconds. This may 
be a solution you would want to look into. Of course, this system was 
not as time critical as the others and a missed event was not 
catastrophic. http://www.arm.com/

I would suggest you look at the sites for http://www.qnx.com and 
http://www.lynuxworks.com. I'm not sure that either of these offers an 
operating system for 8-bit, but they have customer support and sales 
support links on their sites and you may be able to get some answers to 
your questions by contacting them.



Carl.