So far, 3 of the four experiments have run significant work on the Glasgow ARM-farm; the fourth, LHCb, are just getting going. I believe that ATLAS and ALICE were successful. CMS was mostly successful with Monte Carlo data, but ran into some difficulties with detector data: They state the following:
"The CMS physics validation on ARM was mostly successful for Monte Carlo simulated events, but significant differences which are not understood at this time have been found for detector data in several subsystems, including muons, tracking, and the electromagnetic calorimeter. Further investigations will be enabled by using the ARM resources at CNAF during 2024, subject to finding the necessary extra effort, but at this time CMS is not in a position to be able to use ARM processors in production."
Next, I'd like to see if we can get to the point where some fraction of pledges can be made next September using ARM resources, at least for ATLAS and ALICE. To get to this point, we need agreement on what fraction of resources can be pledged. Obviously Analysis workflows have not been (universally) ported or validated so they will want to restrict the level of ARM pledge. My suggestion is 20% - corresponding to a 5-year lifetime and thus an average refresh of 20% of a site.
Would it make sense to have a (pre)-GDB in June for the experiments to present their findings/experiences/concerns and to discuss moving forward? I note that the incentive for Experiments is the reduction of carbon per job; the incentives to sites are (a) reduced carbon generation; (b) reduced power consumption; and (c) more options during procurement to stimulate competition. The downside is increased complexity of running heterogenous infrastructure.