Dean, Saul and I discussed our need for bare metal testing with Jeremy, James, Monty and Clark from the Infrastructure team today.

There are problems to consider and several opportunities.

1) The team is open to working with us on this. 

2) There is benefit to the broader community

a. E.g. Ironic would benefit from BM testing

3) Anyone hosting a bare metal resource has to assume that the incoming images are hostile – they will have root access – so things like platform firmware are potentially vulnerable and would need to be re-flashed after every run.

4) We could build and/or contribute to such a datacenter in the DMZ @ Intel or within the OSU OSL lab. 

a. In either case, someone has to have the job of keeping the systems running and working

5) The existing infra (nodepool) could handle allocating resources out of bare metal clouds

a. We would have to provide “bare metal as a service” software to manage allocation of systems out of the lab

6) There is a robustness issue – if there is only one bare metal cloud – if that cloud has an issue, development stops

a. The current OpenStack infra is based on 8 clouds, if one goes down the jobs are routed to running clouds

b. If we could partner with another interested party (which might be CERN), and had multiple bare metal clouds that would be a more reliable resource for the community.

     brucej