The fundamental ideas

The fundamental ideas of Chapter 4 naturally divide into two groups: those with which we were satisfied, and those with which we were not. The ideas which worked out well were:

• The concept of an abstract machine,
• Capability based protection,
• Processes,
• Layered implementation,
• Uninterpreted I-O devices.

The unfortunate ideas were:

• Mapped address space,
• Distributed system code.

Probably the worst disaster of the project was to attempt to implement a mapped address space on an unsuitable machine (i.e., no mapping hardware). Also, distributed system code turned out to be considerably more difficult to design than we anticipated, leading to some very complicated and not very well understood programs. These problems are more fully discussed in Chapters 20 and 21. The concepts of an abstract machine and capability based protection so permeated our thinking that it is impossible to conceive of the project without them. They provided the essential framework supporting all of our design work. These ideas are explored more fully in Chapter 16. The ECS system was the cleanest realization of these ideas. Unfortunately, as implemented, the computation cost of an ECS system call (virtual instruction) was higher than we anticipated. Chapter 18 explores improvements in the implementation which would have substantially reduced this cost, while Chapter 19 proposes hardware modification to the CPU which would further reduce the cost.