[Planetlab-devel] PLC system discussion (in the hall)

[Stephen Soltesz]

Hello, Larry,

I don't know that I can defend the idea of dropping PCUs, since I wasn't 
motivating the problem of node types originally, nor present for the majority of 
the discussion Wednesday.

To clarify the distinction though, I'm guessing that you're looking at these 
objects in terms of their actual functionality.  And, I think I was looking at 
them more in terms of their representation in the DB or via the API.

My thoughts (and I suspect others that proposed dropping the PCU API calls) come 
from the observation that the PCU representation within the API/db is very 
similar to the representation of a node within the API/db.  And the proposed 
generalization of the generic node as a 'name and collection-of-attributes' fits 
for PCUs as well.  The interpretation follows from the type.

This 'externalization' of the type is what motivated my previous email with the 
diagram.  My thought is that it is important that we recognize the consequence 
of moving types outside of the API, in terms of documentation and homogeneous 
interpretation by our code and third-parties.

 From a functionality point of view, I can understand why this sounds strange. 
"Nodes" are general purpose.  "PCUs" are fixed-function.  "Why generalize this?" 
  The question is good.

Stephen.


Larry Peterson wrote:
> It seems strange to fold PCUs into this. A "node" can always have
> auxiliary pieces -- special cards, adjacent boxes, and so on. These
> seem to be attributes of the node (with node-specific hardware
> responsible for dealing with this sub-piece) rather than visible to
> PLC (and the API).
> 
> Or have I completely misunderstood...
> 
> Larry
>