This is great, TFS. Interesting to read period ATM deployment costs (USD$100K) vs. bank branch rollout costs (USD$1M): I see modern parallels to our space (fully automated contactless food preparation and retail). Also interesting to see frank admission of extreme SPOFs:
There are over 300 banks in Texas on CIRRUS, but they're all connected to a single node. Interesting how ridiculously inertia driven the whole banking industry is, when you compare modern ATM interfaces to statements like
We support withdrawls and balance inquiry from checking, savings and line-of-credit accounts. [...] We're planning to do automatic currency conversion for international transactions. Even on the other side of the world here in China, that basically defines the menu options available today some 40 years later. And the original HQ was Detroit! Oh how the mighty have fallen...
Also amazing: they only targeted 95% uptime (excluding 12 Federal Reserve bank holidays when they were down anyway) and many banks were unable to even reach that level of service! The switch itself was above 98%. This is horrific by today's standards, as was the maximum processing time (45 seconds) and suggested transmission latency allowances (4 seconds). Team size: <1 full time programmer. Code length: 9K lines. TX volume: 50K/month, 4K/day peak.
Further info on TANDEM NONSTOP II hardware which "does not require outages for hardware maintenance" @ https://cs.stanford.edu/people/eroberts/courses/soco/project... ... a few years after this article was written in the subsequent generation TANDEM hardware they added 26MB/sec fiber optic interconnects between the local CPU nodes.