Some relevant background
LTSS (Livermore Timesharing Systen for the 6600 circa 1962) used model 33 teletypes as terminals.
It follwed MIT’s CTSS which used model 28 teletypes.
The disks to which we swapped were large physically and, for their time, had the large capacity of several megabytes.
It took nearly one second to switch from one large job to another.
Head motion time was about 200ms.
The transfer rate was nearly 10 Mb/sec.
Main memory was 128K words of 60 bits each.
We were able to hold parts of more than one job in memory, if those jobs were not too large.
The Scheme
Each command line was terminated by a construct “/ T V” where T and V were decimal integers.
Think value for V and time for T.
Competing requests were sorted by the respective values of V/T.
The highest unserved request ran next.
The request was limited to running T seconds of CPU time.
The user’s account was decremented by the time actually taken times V/T.
V was thus the maximum cost to the user’s account.
You did not run when your account was empty.
You could transfer parts of your account to another user.
Some attention was given to having a production job that was in core and able to run while interactive jobs were swapped in.
I don’t how well that worked.
Problems
Several problems soon became obvious—different problems to different people.
Jobs requiring much less than a second of run time might run and consume nearly one second of the scarce capacity to swap jobs yet pay only in proportion to the small CPU time that they used.
Early Practice
We had planned to elaborate on this scheme as we identified problems but we were thwarted by success.
The machines were a major part of the budget and control over their allocation was a major form of politics, good and bad, at the lab.
The ability to transfer funds remotely was immediately seized upon as an administrative tool with much quicker response than the paper and telephone methods that had gone before.
We had neither coded nor promulgated the discipline that money supply was constant.
Any macro economist would predict what happened next; inflation.
It was easier to create money than take it away from others.
I recall a scheme to automatically trickle money into user accounts at an administered rate.
I do not recall if that was implemented.
Suggestions to change the details of the bidding were resisted by the users of the administrative tools, perhaps for fear of upsetting a working tool.
We were unable to rationalize some of the poor incentive structures that were discovered.
I left the lab not too long afterwards and others may better recall the evolution of the scheme and politics surrounding it.