💾 Archived View for gemini.bortzmeyer.org › rfc-mirror › rfc757.txt captured on 2021-11-30 at 20:18:30.
-=-=-=-=-=-=-
RFC 757
A Suggested Solution to the Naming, Addressing, and Delivery
Problem for ARPAnet Message Systems
Debra P. Deutsch
10 September 1979
Bolt Beranek and Newman
50 Moulton Street
Cambridge, Massachusetts 02138
(617) 491-1850
Preface Page 1
Preface
Unlike many RFCs, this is not a specification of a
soon-to-be-implemented protocol. Instead this is a true request
for comments on the concepts and suggestions found within this
document, written with the hope that its content, and any
discussion which it spurs, will contribute towards the design of
the next generation of computer-based message creation and
delivery systems.
A number of people have made contributions to the form and
content of this document. In particular, I would like to thank
Jerry Burchfiel for his general and technical advice and
encouragement, Bob Thomas for his wisdom about the TIP Login
database and design of a netmail database, Ted Myer for playing
devil's advocate, and Charlotte Mooers for her excellent
editorial assistance.
Debbie Deutsch
RFC 757 September 1979
Introduction Page 2
1. Introduction
The current ARPAnet message handling scheme has evolved from
rather informal, decentralized beginnings. Early developers took
advantage of pre-existing tools -- TECO, FTP -- in order to
implement their first systems. Later, protocols were developed
to codify the conventions already in use. While these
conventions have been able to support an amazing variety and
amount of service, they have a number of shortcomings.
One difficulty is the naming/addressing problem, which deals
with the need both to identify the recipient and to indicate
correctly a delivery point for the message. The current paradigm
is deficient in that it lacks a sharp distinction between the
recipient's name and the recipient's address, which is the
delivery point on the net.
The naming/addressing scheme does not allow users to address
their messages using human names, but instead forces them to
employ designations better designed for machine parsing than
human identification.
Another source of limitations lies in the delivery system,
which is simply an extension of the File Transfer Protocol. The
delivery system is fairly limited in its operation, handling only
simple transactions involving the transfer of a single message to
a single user on the destination host. The ability to bundle
messages and the ability to fan-out messages at the foreign host
would improve the efficiency and usefulness of the system.
An additional drawback to the delivery system is caused, to
some extent, by the addressing scheme. A change in address, or
incorrect address usually causes the delivery system to handle
the message incorrectly. While some hosts support some variety
of a mail forwarding database (MFDB), this solution is at best
inadequate and spotty for providing reliable service to the
network as a whole. Because the same username may belong to
different people at different hosts, ambiguities which may crop
up when messages are incorrectly addressed keep even the best
MFDBs from always being able to do their job.
This proposal envisions a system in which the identity and
address of the recipient are treated as two separate items. A
network database supports a directory service which supplies
correct address information for all recipients. Additionally,
the scheme allows mail delivery to be restricted to authorized
users of the network, should that be a desirable feature.
RFC 757 September 1979
Names and Addresses Page 3
2. Names and Addresses
Today's ARPAnet naming and addressing scheme (as specified in
RFC 733[3]) does not discriminate between the identity of a user
1
and his address . Both are expressed the same way:
USERNAME@HOST. While this should always result in a unique
handle for that user, it has proved to be inadequate in practice.
Users who change the location of their mailboxes, because of
either a change in affiliation or a simple shift in host usage,
also get their names changed. If their old host employs an MFDB
the problem is not too bad. Mail is simply forwarded on to the
new address, slightly delayed. Other less fortunate users who
cannot rely upon an MFDB must notify all their correspondents of
the change in address/name. Any mail addressed to the old
address becomes undeliverable. (An excellent discussion of the
differences between naming, addressing, and routing is found in a
paper by John Shoch [1].)
The desire to use "real" names in the address fields of
messages is also thwarted by the current system. An elaborate
system for using human-compatible vs. machine-interpretable
2
information has been evolved for use in message headers . The
most recent developments indicate that many users would feel
happiest if the real human name could appear;
machine-interpretable information should not intrude too heavily
into the writer's work- and thought-space.
The solution proposed here calls for a total break between the
way a recipient is named or identified and the way in which his
location is specified. Since the ARPAnet is a regulated
environment, a unique (and not necessarily human-readable) ID
could be assigned to each authorized recipient of network mail.
This ID would stay with the user throughout his lifetime on the
network, through changes in address and affiliation.
A network database (which could be derived from the same
database that has been proposed to support TIP login) would
associate each ID with one or more addresses indicating where the
mail for that ID may be delivered. If more than one address were
_______________
1
See, for example, RFC 733's discussion of the semantics of
address fields, in which it is specified that the To: field
"contains the identity of the primary recipients of the message".
2
See the "Syntax of General Addressee Items" section of RFC
733.
RFC 757 September 1979
Names and Addresses Page 4
associated with an ID, that would indicate an ordered preference
in delivery points. The delivery system would attempt delivery
to the first addressee, and, if that failed, try the second, and
3
so on . Most IDs would probably have only one address. Also
associated with each ID would be some information about the ID's
owner: name, postal address, affiliation, phone number, etc.
Rather than being forced to type some awkward character string
in order to name his correspondent, the writer would have to
supply only enough information to allow some process to determine
the unique identity of the recipient. This information might be
the recipient's name or anything else found in the database.
The access to this data would also free the writer from any
need to know the location of the recipient. Once the unique ID
were known, the correct location for delivery would be only a
look-up away.
2.1 A distributed database approach
It is clear that if the network database had only one
instantiation there would be a tremendous contention problem.
All message traffic would be forced to query that one database.
This is extremely undesirable, both in terms of reliability and
speed. It is also clear that requiring each host to maintain a
complete local copy of the entire network database is an
undesirable and unnecessary burden on the hosts.
A better approach would be to build some sophistication into
the local delivery system, and use local mini-databases which are
based upon the contents of a distributed network database. (It
may be redundant and/or partitioned, etc., but is probably not
resident on the local host.) When a local host queries the
network database about an ID (or, in a more costly operation,
asked to supply an ID given enough identification such as name,
etc.) the database may be asked to return all its information on
that ID. At this point the local host can enter all or some of
that information into a locally maintained database of its own.
It will always refer to that database first when looking for a
name or address, only calling the network database if it cannot
find a local entry. Depending upon the desired level of
sophistication of the local message handling programs, additional
information may be added to that database, including, for
_______________
3
Multiple addresses might also be used to indicate that
multiple deliveries are desired.
RFC 757 September 1979
Names and Addresses Page 5
example, nicknames.
The database might be shared by a cluster of hosts (such as
exist at BBN or ISI), or it might be used by only one. Hosts
which originate small amounts of message traffic might rely upon
the network database entirely.
The structure and maintenance of the local databases is left
solely to the local hosts. They may or may not store addresses.
It may be desirable either to garbage collect them, or to let
them grow. The local databases might be linked to smaller, more
specialized databases which are owned by individual users or
groups. These individual databases would be the equivalent of
address books in which users might note special things about
individuals: interests, last time seen, names of associates,
etc. The existence and scope of these databases are not mandated
by this scheme, but it does allow for them.
The same individual databases may be used by message creation
programs in order to determine the recipient's ID from
user-supplied input. For example, a user may address a message
to someone named Nick. The message creation program may
associate "Nick" with an ID, and hand that ID off to the delivery
system, totally removing the matter of address or formal ID from
the user's world.
2.2 Delivery
The delivery operation consists of three parts:
1. Determining the address to which the message must be
sent,
2. Sending the message,
3. Processing by foreign host.
The first step usually means looking up, in either a local or
the network database, the correct address(es) for message
delivery, given the recipient's ID. Should the ID not be known
at the time the message is submitted for delivery, any operation
necessary to determine that ID (such as a call to either the
local or network database) is also performed as part of this
step.
The second step is not too different from what happens today.
The local host establishes a connection to the foreign host. It
is then able to send one or messages to one or more people. The
options are:
- Bulk mail. Several recipients all get the same message.
RFC 757 September 1979
Names and Addresses Page 6
- Bundled mail. Several messages get sent to the same
recipient.
- A combination of the above
- One recipient gets one message.
The foreign host should be able to accept mail for each ID.
The rejection of mail for a given ID by the foreign host would
usually indicate an inconsistency between the sender's local
database and the network database. In this case, the local host
updates its local database from the network database, and
attempts delivery at the "new" host. (This is mail forwarding.)
If a host taken from the network database is found to be
incorrect, there is a problem in the network database, and
appropriate authorities are notified. Thus, address changes
propagate out from the network database only as the out-of-date
information is referenced. This reduces the magnitude of the
local database update problem.
Once the foreign host recognizes the ID(s), the message(s) may
be transmitted to the foreign host. Upon successful
transmission, the job of the local host is done.
The third step requires the foreign host to process the
message(s). This is analogous to what may occur in a mail room.
A foreign host may have to sort the bundled or bulk mail it
receives. In addition, the foreign host might perform internal
or external fan-out functions or other special functions, at the
option of the ID owner.
The implemention and design of possible functions which may be
performed in the mail rooms are neither mandated nor restricted
by this delivery scheme. Since they are too numerous to allow
even a small portion of them to be described here, only a few
examples will be mentioned.
Fan-out functions might include placing messages in multiple
files, sending copies to one or more other users, or
rebroadcasting the messages onto the network. (In that last
case, the foreign host might evaluate an ID list, in much the
same way that the ITS mail repeater broadcasts messages addressed
to certain mailboxes.) Special functions might include automatic
hard-copy creation or reply generation, processing by various
daemons, or any other service found desirable by the host's user
population and administration. The implementation of fan-out
functions is up to the local host, as are any additional
functions which the user population might wish of its local "mail
room". Whatever services are available, the mail room will
distribute the mail to the correct location for each ID.
RFC 757 September 1979
Names and Addresses Page 7
2.2.1 Additional delivery options
It may be desirable to allow mail rooms to accept a username in
place of an ID. Use of a username is a less reliable method of
addressing than use of an ID.
- A username may not be sufficiently unambiguous for
getting an ID and host from the network database.
- Since a recipient's username may change from time to
time, there is a chance that the username supplied by
4
the sender will be incorrect , or that the host may not
recognize it.
Because a recipient's ID does not change with time,
errors such as those caused by username changes cannot
occur if IDs are used. Similarities or ambiguities can
be discovered before delivery occurs, and the sender can
be prompted for additional identifying information about
his intended recipient.
- In an even worse case, a correct username can still
result in an incorrect delivery when it is paired with
an incorrect host or acted upon by a mail forwarding
5
database .
Because unique IDs are unambiguous, the possibility of
such a situation is eliminated by the use of unique IDs.
_______________
4
A particularly insidious source of addressing errors stems
from the inconsistent use of (human) names and initials to
generate usernames. The sender can easily guess his
recipient's username incorrectly by using, or failing to use
a combination of initials and last name. (For example, a
user wishing to address Jim Miller at BBNA and using the
address "Miller@BBNA" will have his message successfully
delivered to Duncan Miller at the same site.)
5
The author has observed a mail forwarding database
redirect messages correctly addressed to one JWalker to
different JWalker at another host.
RFC 757 September 1979
Names and Addresses Page 8
2.2.2 Failures
The case in which the network database is found to be incorrect
has already been discussed. It may make sense to mark the entry
as "possibly in error" and to notify both the network database
6
and the ID owner when such a situation occurs. In this case mail
delivery to the ID's owner will not occur, but this is not too
bad, considering that that is what happens today when a host does
not recognize a username.
One additional failure mode, the loss of the network database
from the net, must be considered, even though a well-designed
distributed network database should be robust enough to almost
rule out this possibility.
If such a failure should occur, the local databases should be
able to handle most of the traffic. What would be lost is the
ability to add new IDs to the network database, the ability to
change hosts for an ID, the ability to update local databases,
and the ability to query the network database. In essence, there
would be a regression to the state we are in today.
A well-administered network database should be backed up
frequently. Should a catastrophic series of hardware failures
remove one or more of the network database's hosts from the net,
the database could be moved elsewhere. Such a change would
entail notification of all hosts on which mail originates.
Software which queries the database should be designed to be able
to easily handle such a move.
_______________
6
Such notification would presumably be by hardcopy mail or
telephone.
RFC 757 September 1979
Relationship to TIP Login database Page 9
3. Relationship to TIP Login database
A number of references to the TIP Login problem and a database
which has been proposed as part of its solution have been made in
this note. A series of working papers [5] written by Bob Thomas,
Paul Santos, and Jack Haverty describe an approach to TIP Login.
In brief, the method is to build and maintain a distributed TIP
Login database, containing information necessary to allow a new
entity called a "login-host" to decide whether or not to grant a
user access to a given TIP, and whether or not to allow the user
to make various modifications to the database itself.
The TIP login database is derived from a "network user data
base", which contains information above and beyond that necessary
to support TIP login. This comprehensive database is designed to
support applications other than TIP Login, either directly or by
means of databases derived from it.
Contained in the TIP Login database are each user's login
string, a list of TIPs the user is authorized to access, the
user's unique ID, his password, and any other "permissions" (in
addition to which TIPs may be accessed). These permissions may
indicate that the user may create, delete, or modify entries in
the database, to assume other user's roles, and to what extent he
may do so. The notion of permissions as developed by Steve
Warshall is discussed in an NSW memo [2].
It seems entirely reasonable to derive a netmail database from
the same comprehensive database that is designed to support TIP
Login. The concept of a unique ID is supported by that database.
Much of the required information for a netmail database is
already included, and the maintenance tools necessary to modify
it seem well-suited for the purpose. The concept of permissions
extends well to the needs of netmail. Permissions specific to
network mail might include, for example, the ability to modify
the delivery host list associated with a given user.
The mechanisms necessary for the maintenance of the
comprehensive network database and its derived databases give us
a netmail database very inexpensively. This proposal takes
advantage of that situation.
RFC 757 September 1979
Relationship to RFC 753 Page 10
4. Relationship to RFC 753
RFC 753 [4] describes an internetwork message delivery system.
Very briefly, the approach is to locate one or more "message
processing modules" (or MPMs) on each network. These MPMs pass
messages across network boundaries, and are also capable of
making deliveries to users on the local network. The document
also details a proposed message format, along the envelope and
letter paradigm. An external "envelope", read by the delivery
system, allows the (unread) message to be correctly routed and
delivered to the proper recipient. Groups of messages passed
between a pair of MPMs are sent together in a "mail bag".
This proposal differs from RFC 753 in that it is primarily
intended to operate within a network or a concatenation of
networks using a common host-host protocol, e.g. TCP. Where RFC
753 addresses the problems of internetwork communication
(differing message formats, complex routing, and correct
identification of the proper recipient), this note concentrates
primarily on what can be done within a single protocol. The two
are not incompatible. While a general internetwork protocol must
provide general methods which can be compatible with different
host-host protocols in different networks, a proposal such as
this one can capitalize on the capabilities, resources, and
policies of a given catenet (catenated network) such as the
ARPAnet/PRnet/SATNET etc.
4.1 Compatibility
The delivery system described in RFC 753 is compatible with the
system outlined here. Let's examine this for each of the three
basic delivery options performed by the MPM. (In the discussion
that follows, "local networks" means a concatenation of networks
using a common host-host protocol, e.g. TCP. "Foreign network"
means some network which uses a different host-host protocol,
e.g. X.25. (See Figure 4-1.)
4.1.1 Outgoing message
4.1.1.1 RFC 753
The sender's process hands a message to the local network MPM.
The message may be destined to an address on the local network or
on a foreign network. In the former case, the MPM performs the
local delivery function (see "Incoming message"). In the latter
case, the MPM passes the message along to another MPM which is
"closer" to the end user.
RFC 757 September 1979
Relationship to RFC 753 Page 11
+---------+ +----------+
| | | |
| RCC-NET | | WIDEBAND | .......
| | | NET | . .
+---------+ | | . MPM .
* * +----------+ .......
+---------+ * * * * ....... |
| | +---------+ . . +---------+
| BBN-NET |***| |__. MPM . ..... | |
| | | ARPANET | ....... . .xxxx| TELENET |
+---------+***| |***********. G . | |
+---------+*** ..... +---------+
* * * * ** .......
+--------+ +-------+ ***..... +-------------+ . .
| | | | . . | |--. MPM .
| SATNET | | PRNET | . G .oooo| DIAL-UP NET | .......
| | | | ..... | |
+--------+ +-------+ +-------------+
"Local Nets", TCP based | "Foreign Nets", other
(direct addressing using IP) | host-host protocols