[HH] Metal as a Service

[Tom Metro]

Federico Lucifredi wrote:
> You can think of MAAS as a PXE server on steroids.

Makes sense.


> IPMI.

I see.

I wonder...adding an IPMI controller - which has to remain up and
running constantly - may end up using as much power as some of the low
power ARM boards that you'd be managing in my proposed scenario.

Does MAAS only work with IPMI? I'm assuming if you power up a node by
some other means, it'll still respond to the net boot request.

But really all you need is an ability to power cycle a specific node.
Eventually someone will hack together a computer controlled power
supply...a bunch of USB jacks, each with its own individually
addressable transistor, all controlled by a Pi or Arduino with an
Ethernet shield.


> - PXE support on ARM hobbist level boards is poor or non-existent...

Hardware limitation or lack of firmware?


> - Ubuntu does not run on R.PI boards...

But that only matters for the controlling node, right? MAAS can be used
to deploy other operating systems to the nodes, no?

Do you need to be able to run Ubuntu on the node at least once to enroll
it in MAAS? (I see the setup menu for Ubuntu Server has an option to
join the machine to a MAAS network.)


> "Managing" to MAAS means controlling a pool of PXE-driven devices
> that were enlisted under its control, and provisioning what is
> installed on them.

I didn't get what actually happened when a node was enrolled. I gather
there is some communication between the node's IPMI controller and the
master node to capture the target node's address? Maybe also capturing
the MAC of the node so it can be mapped to the desired OS boot image?


>  Usually "deployment" or "provisioning" is used in
> this context, rather than managing.

Sure, MAAS is just dealing with the power up/OS boot/power down aspects,
and not management of the node while it is running, like a cloud
management tool would.

No doubt someone will (or has already) integrate MAAS into a cloud
management tool. In some cases it could be used to deploy hypervisors to
the bare metal. On other cases you might transparently manage services
across a mix of VMs and bare metal.


> Hope this helps -Federico

Yup, thanks.

 -Tom