[Info-vax] Real time, was: Re: Volatile, was: Re: yet another sys$qiow question

[Johnny Billquist]

On 2015-08-29 21:14, JF Mezei wrote:
> On 15-08-29 14:47, Simon Clubley wrote:
>
>> It doesn't matter how fast Ethernet is; the root problem is that it
>> doesn't have deterministic behaviour.
>
> While the above is correct, one has to take a "big picture" approach.
>
> Say for instance the military protocol is deterministic, but you can
> only poll once per second.
>
> The ethernet is not deterministic, but the maximum jitter you could
> expect is half a second. So the worse case scenario of ethernet, you are
> still below the response time of 1 second which the military protocol
> provides. (random values to make the point).

Where does it state that the maximum jitter is 0.5s? As far as I know, 
you can have collisions every time you try to send, and you'll back off 
a number of times before the controller gives an error back that you had 
excessive collisions, at which point you have to decide that you might 
want to try again, at which point we might be rather far down the line 
as far as time goes...

Not to mention that there are a lot of times where you want something 
better than just "run within a second from now".

Modern switches do reduce issues somewhat, since you do not see the same 
amount of collisions, but ethernet itself has not changed. Now you 
instead have buffering in the switches to (hopefully) solve the 
collision problem, but which instead might introduce dropped packets. 
You decide which is better.

	Johnny

-- 
Johnny Billquist                  || "I'm on a bus
                                   ||  on a psychedelic trip
email: bqt at softjar.se             ||  Reading murder books
pdp is alive!                     ||  tryin' to stay hip" - B. Idol