Embed Notice

@amszmidt @pmidden

Yes, we had internal clock chips on the 3600.

But they weren't very good! There was significant forward/backward drift in seconds as the days went by. And since those machines were reliable enough that they might not be booted for months, this was a problem.

One day, Doug Dodds (I think?) came up with a "solution". (Quotes because it worked for us, with a lot of lispms, but wouldn't work for a customer site with only 1 or 2.) It was to have the lisp machines all ask each other what time it was, and then set their clocks to the average of the result.

Some time chips were faster, some were slower, but it averaged out *quite* nicely.

I had a nice time explaining the Central Limit Theorem to some colleagues who wanted to know why it worked. :-)