sorting was: The System360 Model 20 Wasn't As Bad As All That 3891

in the late 70s, we were sitting around some drinking hole late one friday night trying to think of something that would help promote online computer use inside the company (for others than strickly professional programmers). we came up with the idea of online telephone book facilities (everybody looked up phone numbers).

the challenge was that the (lookup application) code had to be implemented in less a person week and the infrastructure and aministrative stuff was possibly another person week ... and the ongoing support operation had to be less than one or two person days a month (for all corporate phone books). this included periodic refresh of original source from the original corporate site owners, reformating and redistribution.

the typical online phone number lookup also had to be much faster than reaching for a paper copy (of purely local numbers) on the desk.

so the brain dead was to have the phone book as a sorted file and do a lookup using binary search. this was only border line acceptable, the search was updated to radix search based on measured first two letter frequency.

there are some number of issues with name representation ... and so some rules were created for canonical name representation. then the sort routine ... rather than directly using the name field as the sort key ... created a derived sort key from the name field based on the representation rules (however, all the records in the output file were in their original input format). then of course, the lookup routine had to convert the search argument into canonical format and also dynmacally convert each name field to canonical format as it checked for matches.

later we learned that there had been a corporate hdqtrs proposal for doing a centralized online phone lookup, with dedicated datacenter, 20-30 dedicated people and initial budget of $25m. our KISS solution decimated the corporate hdqtrs proposal.