> "CompuServe had some PDP-10s operating until at least 2007"
This sentence sent me down an internet rabbit hole searching for some sort of citation, as it seems absurd on the face of it. I couldn't really find an original source, but lots of comments and references about how important PDP-10s were to CompuServe. They even used clones after the platform was discontinued. It seems CompuServe kept using and updating their minis until the end.
Can you imagine maintaining a PDP-10 in the mid 2000s? Surreal. I know it's non-trivial to port software off an old computer system, but this is a machine with a clock ticking in microseconds. I can't imagine it'd take more than a week or so to reimplement any sort of business logic it had.
> this is a machine with a clock ticking in microseconds. I can't imagine it'd take more than a week or so to reimplement any sort of business logic it had.
You're kidding right? In the old days people had to do insane things with their code to make it run fast enough. I'm thinking for example of programmers working with then-slow disk and drum storage who had to play elaborate timing tricks with their code based on when they expected data on the disk to rotate into position under the read heads. Just a few lines of assembler code, "ticking in microseconds", could conceal a terrifyingly fragile concoction of code. You could spend hours or days trying to understand it before you would dare to make a change.
Sure if you had an accurate, up to date, human-readable description of the system's "business logic" you'd be off to a flying start to migrate it to a new platform.
But the inalienable truth of old mainframe systems is that such a description no longer exists, if it ever did. The system itself - in COBOL, assembler or whatever - is the true description of the business logic. That's why IBM still sells mainframes compatible with their ancestors from over half a century ago. No-one dares to change many of these systems.
OK. You're right, "weeks" was a severe understatement. But we're talking about 30 years and CompuServe wasn't a bank or nuclear power plant or anything. Extracting the business logic should have been a project that was done and gone by the 2000s.
> I'm thinking for example of programmers working with then-slow disk and drum storage who had to play elaborate timing tricks with their code based on when they expected data on the disk to rotate into position under the read heads.
There was an epic story I once read that revolved around this and it’s going to drive me crazy until I find it.
Edit: Found it! It’s from “The Story of Mel, a Real Programmer”:
> Mel never wrote time-delay loops, either, even when the balky Flexowriter required a delay between output characters to work right. He just located instructions on the drum so each successive one was just past the read head when it was needed; the drum had to execute another complete revolution to find the next instruction.
This makes me think of that algorithmic black box that the IRS has. It is spoken of in reverent tones tinged with dread. I wonder if anyone's tried to throw an AI at it yet to figure out what the heck is really going on in it? And maybe decompile it to render it amenable to modification and recompilation.
They weren't the giant DEC-made PDP-10s, AFAIK. They were System Concept clones of the PDP-10 (SC30, SC40). Relatively small, modern (for the time) semiconductors, SCSI I/O, Ethernet & FDDI network.
There were also PDP-10 clones called TOAD from a company called XKL that were sold into the 80s and maybe 90s.
Both were much, much faster than the original PDP-10s.
This sentence sent me down an internet rabbit hole searching for some sort of citation, as it seems absurd on the face of it. I couldn't really find an original source, but lots of comments and references about how important PDP-10s were to CompuServe. They even used clones after the platform was discontinued. It seems CompuServe kept using and updating their minis until the end.
Can you imagine maintaining a PDP-10 in the mid 2000s? Surreal. I know it's non-trivial to port software off an old computer system, but this is a machine with a clock ticking in microseconds. I can't imagine it'd take more than a week or so to reimplement any sort of business logic it had.