The Andrew file system described in this paper is a location-transparent distributed file system that is planned to span more than 5,000 workstations. Such a scale affects performance and complicates system operation. The authors discuss several versions of the prototype and the observed performance of each. They conclude that the existing version could easily grow to handle 500–700 workstations but additional work will be necessary to meet the design goal.
The system described is implemented in BSD UNIX 4.2. The basic concept is to move entire files to a local cache as needed rather than involve the file servers in actual I/O. Two versions with drastically different reactions to scale-up are described. The authors conclude with a statement of further revisions they have done that simplify the establishment of user disk quotas, ease volume maintenance, and assist in backing the system up.
I was quite impressed that, although I am not familiar with BSD UNIX, the paper, which goes into considerable depth, was understandable without constant reference to a manual. This quality of technical writing deserves to be praised (and repeated). The paper touches on a phenomenon that is very popular in the modern computing world (distributed file systems) and one that needs considerable study before being implemented. This study is not just applicable to the academic (university or research) environment but also to many industrial environments where file sharing is needed.