Text of original e-mail message of January 6, 2004, to Brian Hayes, later edited and published as "Letter to the Editors: Past Imperfect"", in American Scientist 92, 2 (March-April 2004), p 110.


Mr. Hayes,

I have long enjoyed your Computing Science columns in American Scientist.  In your January-February 2004 column, you make many interesting points worth pondering, but the paragraph quoted above contains a significant historical error that should be corrected before it propagates any further.

The Unix system contributed quite a number of good ideas, and it was a superb piece of engineering under tight constraints, choosing a useful and effective set of functions that could be implemented well in a small computer.

And your characterization of the hierarchical file system as both an important innovation and desperately in need of a better idea seems appropriate.  But it is not an example of a feature introduced by Unix.

Dennis Ritchie and Ken Thompson participated for Bell Labs in the design and implementation of Multics, a joint project among M.I.T., General Electric, and Bell Labs.  One of the many features in Multics was a hierarchical file system of unlimited depth.  That feature in turn was an outgrowth of experience with a strictly two-level hierarchical file system in the Compatible Time-Sharing System.  CTSS started life with one flat directory per user, but around 1964 its file system was upgraded to place all user directories in a master file directory.  This limited hierarchy was visible primarily to the system itself, but it was so evidently useful that when planning for Multics began in 1964, a user-navigable hierarchy was a top priority.  And it was an operational feature in the first, limping version of Multics in 1967.

Between 1960 and 1967 several organizations undertook projects to create time-sharing systems.  Because it was apparent by 1964 that it might be a good idea, it is entirely possible that one of those other projects had a hierarchical file system running before Multics, but I don't recall seeing anyone else make a clear claim of priority.

Ritchie and Thompson made significant contributions to the design of Multics, in its compilers and its I/O system, respectively.  And they went on to make many other significant innovations in the design of Unix, examples including the C language and pipes.  But many of the features of Unix, including the hierarchical file system, are of greater antiquity.

                                  Jerry Saltzer