We had vaguely imagined from the beginning of Tymnet the real-time automation of new circuits. The SDS 940 was the largest machine we had available (192 KB) and so we used it to dynamically setup circuits at the beginning of each login session. The new program was called the supervisor.
We converted a small fraction of Tymnet, perhaps 20 nodes around Cupertino, to the new system. With the new system any dialup port would be able to reach any timesharing host. We went thru the normal debugging stages which were eased since the nodes were all local. There were manifest difficulties in arranging yet larger test networks. The flexibility of switched circuits was badly needed. The cost of the switchover included manually reloading a few hundred nodes in many cities. We finally decided to bite the bullet and convert the whole network. One Friday evening as the timesharing load waned, we took down the whole network by beginning to load the new Tymnet node code into each of the many nodes.
The 940 struggled valiantly reporting the number of nodes whose connections it had discerned. We waited and would take occasional steps to help it along by adjusting heuristic parameters. I am not of much use after midnight and I went home. Laroy (Tymes) stayed on. Early the next morning I came in and found all of Tymnet up with customers. I assumed that Laroy, who had written all the code, had found and fixed the problems. An hour or so later Laroy came in and assumed that I had found the problems that morning. The supervisor had not been able to come up while we were still there and meddling. After Laroy had finally left the supervisor did its methodical thing and had mapped and taken over the whole net.
Of course other bugs were found and improvements were made but Tymnet remained supervised from that point on.