Sandy Fraser was kind enough to share some papers from his archive that give further
background to early networking at Bell Labs. One of these is about the “File Store”.
For context I refer to an article that Sandy wrote back in 1975 and describes the network
setup at Murray Hill at that time:
https://drive.google.com/file/d/1_kg4CEsbGucsU8-jxi0ptfUdsznWcKWm/view?usp=…
In the figure and legenda at the bottom of page 52/53, the "File Store" is item
10.
The File Store paper itself is (temporarily) here:
https://drive.google.com/file/d/1AhLmjJcHXFtfoIUlfvl0bzbB0zSPzpSq/view?usp=…
The paper has a very interesting introduction: “Five machines currently use the file
store. Three of them use it as if were a peripheral device not part of their main backing
store. In these cases there exists programs to transmit complete files between the user’s
machine and the file store."
This was known and Noel Chiappa found the source for the main transfer program “nfs” (this
program is also mentioned in Doug McIllroy’s manual compendium):
https://chiselapp.com/user/pnr/repository/Spider/tree?ci=tip
The introduction continues: “In the other two cases the file store is treated as an
extension of the user machine’s backing store. Once a user has opened a file his program
does reads, writes, and seeks without being aware of the file’s actual location.”
This -- to me at least -- is a new fact and as such it would predate various other
projects for a distributed Unix file system (the paper is dated December 1974).
Unfortunately, the paper is short on how the integration was achieved.
On one hand the work may have been related to "Peripheral Unix” as developed by Heinz
Lycklama
(
https://www.tuhs.org/Archive/Documentation/TechReports/Heinz_Tech_Memos/TM-…)
at the same time -- the memos are dated just a month apart. In essence the approach is to
forward system calls over the network 1).
Another possibility is that it worked much like the 1979 “RIDE” system
(
https://www.computer.org/csdl/proceedings-article/cmpsac/1979/00762533/12Om…). Here
there is a modified C library that recognises certain path names and maps these to file
server calls.
A third possibility is that it was a precursor to the work on distributed Unix by Luderer
et al. in 1980/81 (
https://dl.acm.org/doi/abs/10.1145/1067627.806604). Here file system
calls are redirected at the kernel level using the concept of a remote/forwarding inode.
The File Store paper mentions that the server is a modified Unix. At first glance it would
seem that the modifications are modest, with the file system partly rewritten to account
for storage usage, and an automatic backup feature added.
I am much interested in any recollections, insights and materials about these topics.
Many thanks in advance,
Paul
Note 1)
The tech report on the “high speed serial loop” (the Weller loop) has not surfaced yet,
but the document for the Glance terminal gives a quick, high level overview on page 3/4:
https://www.tuhs.org/Archive/Documentation/TechReports/Heinz_Tech_Memos/TM-…
The recent 516 documents include a detailed description of how it worked:
https://www.tuhs.org/Archive/Documentation/Other/516-TSS/516-10-11-12-Ring-…