The current policy, which was established when Lorenzo Foá was director, is as follows:

Institutes may have their equipment installed in the computer room provided that it follows precisely the current CERN standards (preferably is purchased using the current CERN tender) and is managed as a standard system by IT-FIO. The latter condition means that the institute gets access to the capacity owned by it, but not necessarily on the specific boxes they own. The equipment will be declared obsolete at the same time as equivalent CERN owned equipment is retired.

IT and the concerned experiment must both agree with this (i.e. this is a privilege not a right). In addition it is the experiment, not the institute, that agrees with IT how to configure scheduling of the resources. In other words, CERN sees this as (additional) capacity allocated to the experiment, and it is up to the experiment to agree on any special internal allocations. 

The funding provided by the institute must cover the price of the box, the cost of the installation (racks, cabling, network concentrators, manpower for the installation) and a lump sum for three years of system administration. The current cost for these overheads is CHF 750 for each CPU server.

One problem is that the capacity of CERN's Computer Centre is limited, mainly by the available power and cooling capacity. Current estimates are that only a few hundred additional private CPU nodes could be accomodated on top of the equipment planned for the Tier-0 and the CERN Analysis Facility. On the other hand, as suggested by my previous mail to the PEB, the funding is not yet assured to meet the planned capacity!

I would like to discuss this at the next PEB meeting, but in view of the difficulty of finding dates during September, I would appreciate if we could have a first discussion by email.

It is understood of course that externally funded equipment installed in CERN's Computer Centre which forms part of the planned capacity for the experiment is reported to the C-RRB as a contribution to LCG  and is not subject to these constraints.