[e2e] Nature mag, DARPA, and the Internet

David P. Reed dpreed at reed.com
Sun Aug 10 08:51:30 PDT 2003

Multics had what was essentially an extensible, almost-object-oriented file 
system.  That is, one could create objects that were accessible only to the 
methods associated with the objects (enforced by the ring/domain protection 

Multics mailboxes were extended filesystem objects that contained messages 
(typically email, but the mailboxes could also be used for interprocess 
communication).   One could "append" a message to a mailbox, and read 
messages and delete them, through the methods of the mailbox system.

At the lowest level, mailboxes were stored in files, but normal user APIs, 
and the mail delivery system itself, could not access those files as files.

Access to the files were controlled by Access Control Lists, with separate 
per-user/group entries that had permission bits for Append, Read, Write, 
Delete, ...

Normally, as I recall, intra-machine messages were delivered directly into 
the mailbox of the recipient by the sender.  The reason for this is that 
one could use the ACL to specify who could access a mailbox for what.   For 
example, you might let yourself have all access.   Your personal assistant 
could have Read access, but not Delete.   Everybody might have Append 
access, except for people you didn't like.

I don't remember what kinds of inter-system message delivery mechanism we 
had prior to the ARPANET mail protocols.   I recall that we may have, but 
they were certainly NOT general, because we didn't have a way to represent 
general addresses outside the file system.   However, local mailboxes could 
be used as standins for interoperation with remote ones, and it was 
generally understood how to do this.

I believe that the major contribution of ARPANET email was in standardizing 
addressing and contents across systems.

At 03:00 PM 8/9/2003 -0400, John Day wrote:
>At 11:12 -0700 8/9/03, Bob Braden wrote:
>>   *>
>>   *> Mea culpa on the mouse. Again the DARPA PR engine has failed us. NLS
>>   *> might be a good ref for email too, for that matter.
>>   *>
>>   *> Joe
>>   *>
>>No, email originated under Multics and then at BBN -- both were
>>DARPA funded.
>Are you sure about Multics?  The reason I ask is that I distinctly 
>remember that to implement email in FTP Multics added a new access control 
>parameter: append.  So the parameters were read, write and append.  In 
>other words, you had permission to add stuff to the end of a file, but not 
>to write any where in the file.  Now that said, I find it hard to believe 
>that Multics didn't have some sort of mail function prior to 
>1973.  Perhaps each piece of mail was a separate file?  Can one of the old 
>Multics hackers whose memory is better than mine clarify the situation.
>Take care,

More information about the end2end-interest mailing list