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

[Larry Peterson]

Hmm. From an architectural perspective, either you run a CM (or someone runs
it on your behalf), or you are merely an implementation detail. I don't buy the
spectrum argument. You can't be half pregnant.

An IXP or OpenFlow router might be represented by a proxy CM that runs on
another box, but that's ok. A PCU is another matter. I can understand why PLC
needs to keep track of PCUs as a matter of practical interest, but calling them
"nodes" (just because the DB and API know about nodes) seems wrong.

Don't get me wrong. I try to stay squarely on the purity/pragmatist
fence. It might
be ok to piggyback PCUs on some existing mechanism -- for the sake of
conserving mechanism -- but I'd be hesitant to perturb that mechanism in any
way to support this piggybacking.

Larry

On Fri, May 16, 2008 at 3:15 PM, Marc E. Fiuczynski
<mef at cs.princeton.edu> wrote:
> Stephen and Larry,
>
> Stephen, you basically captured the reason for making PCU just another node
> type: in terms of representation and the API it basically is a kind of node
> (albeit very limited).   Basically Thierry and others felt that it was an
> unnecessary pain to maintain such information separately.
>
> Larry, the goal of our hallway discussion was the definition of an abstract
> node in order to make it trivial to support other types of nodes.
>  Specifically the goal was to uniformly capture the full spectrum of nodes
> that play an integral role in a planetlab network.  At one end of the
> spectrum we've got the full fledged nodes that support slices etc. and at
> the other end fixed-function devices that themselves do not (yet) honor the
> "component manager" interface (e.g., PCUs).  What's more interesting will be
> the nodes in the middle that support some variant of the "component manager"
> interface (e.g., wifi mesh nodes, ixp boards, smart ethernet switches,
> supercharged planetlab nodes, etc.).
>
> Marc
>
> _______________________________________________
> Devel mailing list
> Devel at lists.planet-lab.org
> https://lists.planet-lab.org/mailman/listinfo/devel
>
>