Tom O’Rourke was a salesman for GE’s very early timesharing service. He saw the potential in that service industry and left, with Dave Schmidt, to form their own company, planning to buy the hardware-software systems from GE. GE decided not to sell these systems, at least to competitors. Just about that time the Genie project at Berkeley was beginning to show signs of being useful. A group there had started with an SDS 930 computer with several attached model 33 Teletypes. Its core memory had 64K words of 24 bits each. Access to core was 1.75 microseconds. Cheap magnetic tapes, paper tape equipment, and a line printer completed the configuration. Butler Lampson’s paragraph from this page is a good short description of the origins of that project.
With this small amount of memory only about two users could be served responsively. User pages could be “swapped out” to magnetic tape and of course it was planned to get some sort of faster access storage for swapping.
The Berkeley group had documented the hardware changes that they had made to the machine very well and SDS agreed to deliver a new series of machine, the SDS 940, which incorporated just those changes. SDS also arranged to interface a Data Products disk and some fixed head Vermont Drums to the system to make the system better suited for timesharing.
Tymshare took the software from Berkeley, several releases, actually, and began to develop it for commercial use. We adapted the file system to the disk and the swapping logic to the drums. Butler Lampson wrote this about the Berkeley software and the modifications that Berkeley had made to the SDS 930 about that time.
Our first Data Products disk held 16 MB. There were 8 platters, each several feet in diameter. We upgraded this to 64 MB before we went commercial.
There were 10 pages (20K instructions) of privileged code and data. This was not swapped. All else was swapped. The “exec” sat between the privileged code and the user code and was about again as large. The exec provided file directory logic and services like the modern Unix shell. The exec was not user replaceable. The system could handle about 12 users well upon first commercial service and later nearly 40 users.
Harvard had contracted with SDS to purchase a 940 and software suitable for timesharing. SDS delivered the machine but had not improved the software from Berkeley to the extent that we had. They tried to buy it from at least 2 other 940 users, but even after weeks of effort, all failed the Harvard tests. We traveled to Harvard, demonstrated our software and passed the tests in 2 days. We thereby sold a software license to SDS (for $5,000) with which SDS fulfilled its contract with Harvard, and sold a million dollar machine.
We had installed several machines in Palo Alto when we opened an office in Ingelwood, just north of Los Angeles, where we installed another 940. Customers had to place a phone call to our computer to use it and the long distance charges were significant for some. We developed Tymnet initially to extend our geographic reach.