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

[Larry Peterson]

Agreed. We have to do inventory tracking. I'm only questioning
how much we stretch our implementation to simultaneously do
inventory tracking and present a tidy interface for user-visible
components. Maybe there's little or no compromise to do both.
On the other hand, I start to worry when I hear "API changes
will be necessary."

Larry


On Fri, May 16, 2008 at 4:15 PM, Marc E. Fiuczynski
<mef at cs.princeton.edu> wrote:
> Just to be clear, I am not trying to elevate a PCU to a GENI-style
> component.  However, as part of the implementation details, there is a fair
> amount of inventory tracking etc. that is needed to build a functioning
> system.  PCUs are slightly more complex than just "on/off".  They have their
> own hostname & ip address (need to be named), password (authentication /
> security), different methods of interacting with them, generally need to be
> managed/monitored by something to ensure that they are operational, AND may
> either manage power to one or more of the "real" components (involves naming
> the components -- i.e., their relationship with each other) .  There is
> sufficient overlap and intermixing at this level with a "component" in terms
> of inventory tracking that it seems easiest / best to build a single
> solution for this.
>
> In my opinion, there is something architecturally clean if the whole system
> can be defined uniformly vs. having some wart glued to its side.  Being able
> to do so uniformly will hopefully help us at some point in the future make
> the claim that we could move *everything* to the next/new generation of the
> Internet (in terms of naming, management, monitoring, etc.).   Otherwise
> people will claim that one could not bring up a new PlanetLab/GENI system
> without relying on DNS, IPv4, etc.  We are now where close to this, but we
> wont progress to this by sweeping the implementation details under the rug.
>
> Marc
>
>
> On May 16, 2008, at 4:00 PM, Larry Peterson wrote:
>>
>> I'm, sorry. I agree not all nodes will support all capabilities... some
>> will
>> be slicable and others not. Some programmable and others not. But a
>> PCU seems like a different beast entirely. Like keeping a separate DB
>> entry for the USB key. I would never ask a CM (or CM proxy) to do any
>> architecturally-visible thing to a PCU. So the litmus test for me is if
>> there is every going to be a CM for the device.
>>
>> I realize we're talking abstractly here, and there is the matter of
>> tracking
>> PCU information. Why isn't this a "field" associated with the node?
>>
>> Larry
>
> _______________________________________________
> Devel mailing list
> Devel at lists.planet-lab.org
> https://lists.planet-lab.org/mailman/listinfo/devel
>
>