Re: Protocol Definition

[John Day <[e-mail]>]

At 6:39 -0800 2012/01/09, Dave CROCKER wrote:
>On 1/8/2012 12:03 AM, t.petch wrote:
>>I agree that a message is not the right word, but I think that protocol is:-)
>
>There is a specific distinction that is intended by having two 
>different words:  description vs. operation.
>
>A program is a description of behavior.  A process is the operation 
>of the description.  It is the behavioral performance.
>
>Protocol refers to the description of an interaction.  The term 
>being explored is for the operation of that description. It is the 
>interaction.
>

Agreed.

>>For the abstract side of networking, I would use the same 
>>terminology as I would
>>use for a 'program'.
>
>If you mean that you would say 'process' for both, that does have 
>the appeal of familiarity, but the problem of overloading.  Worse, 
>I'd fear that it encourages a failure to appreciate the differences 
>between networking architecture and software implementation.  Since 
>this failure is quite prevalent, I suggest we not encourage it.

Well, pretty close.  There is a copious literature on the formal 
description and verification of protocols beginning in the 70s.

There are two major issues that is not normally found in defining a 
"program":  1) is specifying the asynchrony, ensuring no races, and 
2) keeping the specification implementation independent.  One does 
not want the specification to unnecessarily constrain the 
implementation strategy.  The general rule is Day's First Rule of 
Architecture:  Anything you can get away with that is undetectable 
from the outside is legal.  Or when it comes to implementation sleaze 
the architecture.

We have seen the problems of code monoculture or just assuming the 
other guys knew how to code.

Take care,
John