Large

The system was large in a number of ways. First, it occupied a large amount of central memory, thus reducing the available field length for the user. The ECS system, together with I-O buffers, required about 7K words (1K 1000 base 10). The process descriptor together with the subprocesses in the path between the user subprocess and the root required another 4K words. On our 32K machine, this left about 21K words for the user. Furthermore, a user of the SCOPE simulator was penalized another 1K words, leaving him about 20K words. This last can be compared with the real SCOPE operating system, as run at U.C. Berkeley campus computer center in 1971, which occupied 12K words of C.M. On a 32K machine this also would provide a single user with a maximum of 20K words of central. Thus, in some absolute sense, the SCOPE system preempted as much CM space as our system. However, the SCOPE system was run on a machine with 64K of CM, while we had only 32K. Second, the system has a fixed overhead in ECS of 140K words. (Almost half of the available ECS.) This overhead was composed of system code and tables. Most of the space seemed to be occupied by the disk system, but we never had a chance to do a detailed accounting. It was fairly clear that the ECS system accounted for a fairly small fraction of this overhead, and most was due to higher levels of the system. Third, there was a per process overhead in ECS of about 10K words. This figure was better understood than the fixed ECS system overhead and was expected to decrease. It was composed of two major parts:

i)

about 3K was consumed by the local C-lists and storage for the 8 system subprocesses, along with the space necessary to define the process structure and provide a subprocess call stack,

ii)

about 7K was used to provide an ECS image of portions of disk files attached by the process.

It is probable that the implementation of process swapping by the disk system would have reduced this per process ECS overhead to around 3K (only ``swapped-in'' processes would require the 7K for ECS images of disk files). Further reductions would have required a redesign of some portions of the system. Even without process swapping, various developments under way at the termination of the project would have reduced the overhead, possibly by as much as 4K, leaving a 6K overhead. Thus, since only 300K of the 500K ECS was available for the system (the rest was dedicated to the computer center's batch system), at most 16 user processes could exist, even if they were idle. Process swapping would increase this to around 50.