This is based on notes from a discussion on ~4/26:

• We need to consider both inter- and intra-organizational cluster federation
• There is the notion of "ubernetes" – offloading work to AWS. For example, if we find that AWS is cheaper for some functions. Development would be done on NDS resources ("the case of the 6K bug) and deployed to AWS.
• XSEDE resources have specialized characteristics (GPUS, etc)
• Factors to consider include resources, cost, elasticity, SLAs
• Inter-organizational federations between NCSA/TACC/SDSC/PSC. Some groups will run their own.
• We want installations to "phone home" to keep the community close
• We want to be able to move deployments to the resources that make the most sense