Physical Infrastructure Failures Are Exposing Software Assumptions

Two threads today forced engineers to think about physical reality. The AWS Middle East data loss story raised the question of what happens when a cloud provider's physical buildings are destroyed. The engineering explainer on the US Strategic Petroleum Reserve, with its salt dome caverns and ancient monitoring systems, prompted comments about how the gap between physical infrastructure and software assumptions is underappreciated.

The backups thread added a sharp practical edge. The main insight from that discussion: backups have no value, only restores do, and almost nobody tests restores. The thread on the 32-year-old Telnet bug showed that old code carries assumptions from its era that modern infrastructure has long since invalidated.

The pattern across these is that software people systematically underestimate the physical and temporal assumptions baked into infrastructure. Cloud SLAs assume peacetime. Backup systems assume someone will run the restore. Old network code assumes memory is plentiful and input is trusted.