max CPU load recommended

max CPU load recommended

[Gerd]

Sorry, this not an Ada related question, but maybe you could answer it
best:

What is a recommended value for maximum CPU load for safety systems?
Are there any recommendations within 61508 or 26262?

Re: max CPU load recommended

[Vinzent Hoefler]

Gerd wrote:

> Sorry, this not an Ada related question, but maybe you could answer it
> best:
>
> What is a recommended value for maximum CPU load for safety systems?

69.3%, according to the formula. ;)

See <http://en.wikipedia.org/wiki/Rate-monotonic_scheduling>.


Vinzent.

-- 
The most likely way for the world to be destroyed, most experts agree,
is by accident. That's where we come in; we're computer professionals.
We cause accidents.
     -- Nathaniel Borenstein

Re: max CPU load recommended

[Jeffrey Carter]

On 06/26/2012 03:14 PM, Vinzent Hoefler wrote:
> Gerd wrote:
>>
>> What is a recommended value for maximum CPU load for safety systems?
>
> 69.3%, according to the formula. ;)

Most of the critical embedded systems I've worked on said 50%, to allow for 
future enhancements. And none of them ever used that reserved 50%; when they 
came to do enhancements, they upgraded the processor(s), and required that we 
keep 50% for future enhancements.

-- 
Jeff Carter
"People called Romanes, they go the house?"
Monty Python's Life of Brian
79



--- Posted via news://freenews.netfront.net/ - Complaints to news@netfront.net ---

Re: max CPU load recommended

[Niklas Holsti]

On 12-06-25 12:53 , Gerd wrote:
> Sorry, this not an Ada related question, but maybe you could answer it
> best:
>
> What is a recommended value for maximum CPU load for safety systems?
> Are there any recommendations within 61508 or 26262?

I don't know what those standards recommend, but the idea of a "safe" 
CPU load is wrong, IMO. The actual CPU load varies with time; when the 
CPU is processing a task, its load is 100%; when it is waiting for some 
task to become ready, it is 0%. Any measured "CPU load" figure is an 
average over some time period and cannot be used to prove that the 
system meets its deadlines within that time period.

For example, if the system is expected to respond in 1 ms to an 
interrupt that occurs about twice per second, the average CPU load over 
one second can be quite small even if the response takes 10 ms instead 
of 1 ms.

You should rather compute the margins for all deadlines and put a limit 
there, for example 50% margin, or some absolute time margin.

Some rigorous schedulability analysis methods, like RMA, give predicted 
CPU loads and ensure that if this predicted load is below some limit 
then all deadlines will be met. Other (more flexible) analysis methods 
focus on response times rather than CPU loads.

-- 
Niklas Holsti
Tidorum Ltd
niklas holsti tidorum fi
       .      @       .