Using the Cache to Change the Width of Memory 375

when amdahl was starting his mainframe clone company in the early 70s, he gave a talk at mit. he was asked about how he got venture funding. he said something about his business plan pointing out that there was already over $100b spent on 360 applications and that even if ibm walked away from 360 architecture at that moment (could be construed as a thinly vieled reference to the radically different FS architecture being worked on), that customers would still be buying 360 processors 30 years later (i.e. 2000).

the issue was that a large number of applications are meeting requirements and the cost of rewrite as well as possibly opportunity lost cost (putting scarce resources to work on rewriting a working application rather than getting out something brand new) wasn't less than any expected cost savings.

in some cases, there was significant risk issue also considered ... i.e. the current implementations are known to do the job ... there have been numerious mainframe "modernization" (rewirte) projects where hundreds of millions (and even billions) have been spent ... and the projects failed.

misc. past FS (future system) posts:

in the late 60s, i saw a case of a university administrative payroll application appear to have a problem. It had started out as a 407 plug-board application ... which got translated by something (? I don't remember the details), which got translated to 709 cobol, which got translated to 360 cobol. at the end of the program it was still outputting emulated 407 sense settings on the printer. one day the operators notice that the 407 values were differen't than they had been. everything was stopped and the whole datacenter was put on hold while they tried to contact somebody in administrative dataprocessing to see what might need to happen (did it fail and payroll would have to be run again?). They finally found somebody, but couldn't find anybody that knew what the 407 values were supposed to indicate ... so the decision was to run the program again ... and if they came out the same, buttume everything was ok.

--