Re-engineering SAM or Changing the Engine in the Train While it is Running

In the last few years at Fermilab we re-architect-ed our SAM1^[1] family of data catalog and file transfer tools - including major changes - while continuing to transfer over 1 Pb/month of data to multiple existing experiments and bring new experiments on board. This work was done with less than 3 FTE-years of effort, and the changes made include major ones, such as changing interprocess communication protocols, migrating database back-ends, removing and replacing major components, and supporting new file delivery methods. This paper will summarize the approaches we have used to do this, including using design patterns like the Facade, Adapter, and Command patterns, and assisting experiments one at a time with client migration. This process has allowed us to modernize our infrastructure with reasonable costs in both calendar time and developer effort, while continuing to provide the operating service to our customers with minimal interruptions.