RFC 33






Network Working Group                                         S. Crocker
Request for Comments: 33                                            UCLA
                                                                 S. Carr
                                                      University of Utah
                                                                 V. Cerf
                                                                    UCLA
                                                        12 February 1970


                         New HOST-HOST Protocol

   Attached is a copy of the paper to be presented at the SJCC on the
   HOST-HOST Protocol.  It indicates many changes from the old protocol
   in NWG/RFC 11; these changes resulted from the network meeting on
   December 8, 1969.  The attached document does not contain enough
   information to write a NCP, and I will send out another memo or so
   shortly.  Responses to this memo are solicited, either as NWG/RFC's
   or personal notes to me.


                     HOST-HOST Communication Protocol
                           in the ARPA Network*

   by C. Stephen Carr
   University of Utah
   Salt Lake City, Utah

   and

   by Stephen D. Crocker
   University of California
   Los Angeles, California

   and

   by Vinton G. Cerf
   University of California
   Los Angeles, California

   *This research was sponsored by the Advanced Research Projects
   Agency, Department of Defense, under contracts AF30(602)-4277 and
   DAHC15-69-C-0825.

INTRODUCTION

   The Advanced Research Projects Agency (ARPA) Computer Network
   (hereafter referred to as the "ARPA network") is one of the most
   ambitious computer networks attempted to date.  [1]  The types of



Crocker, et. al.                                                [Page 1]

RFC 33                   New HOST-HOST Protocol         12 February 1970


   machines and operating systems involved in the network vary widely.
   For example, the computers at the first four sites are an XDS 940
   (Stanford Research Institute), an IBM 360/75 (University of
   California, Santa Barbara), an XDS SIGMA-7 (University of California,
   Los Angeles), and a DEC PDP-10 (University of Utah).  The only
   commonality among the network membership is the use of highly
   interactive time-sharing systems; but, of course, these are all
   different in external appearance and implementation.  Furthermore, no
   one node is in control of the network.  This has insured reliability
   but complicates the software.

   Of the networks which have reached the operational phase and been
   reported in the literature, none have involved the variety of
   computers and operating systems found in the ARPA network.  For
   example, the Carnegie-Mellon, Princeton, IBM network consists of
   360/67's with identical software. [2]  Load sharing among identical
   batch machines was commonplace at North American Rockwell Corporation
   in the early 1960's.  Therefore, the implementers of the present
   network have been only slightly influenced by earlier network
   attempts.

   However, early time-sharing studies at the University of California
   at Berkeley, MIT, Lincoln Laboratory, and System Development
   Corporation (all ARPAA sponsored) have had considerable influence on
   the design of the network.  In some sense, the ARPA network of time-
   shared computers is a natural extension of earlier time-sharing
   concepts.

   The network is seen as a set of data entry and exit points into which
   individual computers insert messages destined for another (or the
   same) computer, and from which such messages emerge.  The format of
   such messages and the operation of the network was specified by the
   network contractor (BB&N) and it became the responsibility of
   representatives of the various computer sites to impose such
   additional constraints and provide such protocol as necessary for
   users at one site to use resources at foreign sites.  This paper
   details the decisions that have been made and the considerations
   behind these decisions.

   Several people deserve acknowledgement in this effort.  J. Rulifson
   and W. Duvall of SRI participated in the early design effort of the
   protocol and in the discussions of NIL.  G. Deloche of Thompson-CSF
   participated in the design effort while he was at UCLA and provided
   considerable documentation.  J. Curry of Utah and P. Rovner of
   Lincoln Laboratory reviewed the early design and NIL.  W. Crowther of
   Bolt, Beranek and Newman, contributed the idea of a virtual net.  The
   BB&N staff provided substantial assistance and guidance while
   delivering the network.



Crocker, et. al.                                                [Page 2]

RFC 33                   New HOST-HOST Protocol         12 February 1970


   We have found that, in the process of connecting machines and
   operating systems together, a great deal of rapport has been
   established between personnel at the various network node sites.  The
   resulting mixture of ideas, discussions, disagreements, and
   resolutions has been highly refreshing and beneficial to all
   involved, and we regard the human interaction as a valuable by-
   product of the main effect.

THE NETWORK AS SEEN BY THE HOSTS

   Before going on to discuss operating system communication protocol,
   some definitions are needed.

      A HOST is a computer system which is a part of the network,

      An IMP (Interface Message Processor) is a Honeywell DDP-516
      computer which interfaces with up to four HOSTs at a particular
      site, and allows HOSTs access into the network.  The configuration
      of the initial four-HOST network is given in figure 1.  The IMPs
      from a store-and-forward communications network.  A companion
      paper in these proceedings covers the IMPs in some detail. [3]

   A message is a bit stream less than 8096 bits long which is given to
   an IMP by a HOST for transmission to another HOST.  The first 32 bits
   of the message are the leader.  The leader contains the following
   information:

      (a) HOST
      (b) Message Type
      (c) Flags
      (d) Link Number

   When a message is transmitted from a HOST to its IMP, the HOST field
   of the leader names the receiving HOST.  When the message arrives at
   the receiving HOST, the HOST field names the sending HOST.

   Only two message types are of concern in this paper.  Regular
   messages are generated by a HOST and sent to its IMP for transmission
   to a foreign HOST.  The other message type of interest is a RFNM
   (Request-for-Next-Message).  RFNM's are explained in conjunction with
   links.

   The flag field of the leader controls special cases not of concern
   here.







Crocker, et. al.                                                [Page 3]

RFC 33                   New HOST-HOST Protocol         12 February 1970


   The link number identifies over which of 256 logical paths (links)
   between the sending HOST and the receiving HOST the message will be
   sent.  Each link is unidirectional and is controlled by the network
   so that no more than one message at a time may be sent over it.  This
   control is implemented using RFNM messages.  After a sending HOST has
   sent a message to a receiving HOST over a particular link, the
   sending HOST is prohibited from sending another message over that
   same link until the sending HOST receives a RFMN.  The RFNM is
   generated by the IMP connected to the receiving HOST, and the RFNM is
   sent back to the sending HOST after the message has entered the
   receiving HOST.  It is important to remember that there are 356 links
   in each direction and that no relationship among these is imposed by
   the network.

   The purpose of the link and RFMN mechanism is to prohibit individual
   users from overloading an IMP or a HOST.  Implicit in this purpose is
   the assumption that a user does not use multiple links to achieve a
   wide band, and to a large extent the HOST-HOST protocol cooperates
   with this assumption.  An even more basic assumption, of course, is
   that the network's load comes from some users transmitting sequences
   of messages rather than many users transmitting single messages
   coincidently.

   In order to delimit the length of the message, and to make it easier
   for HOSTs of differing word lengths to communicate, the following
   formatting procedure is used.  When a HOST prepares a message for
   output, it creates a 32-bit leader.  Following the leader is a binary
   string, called marking, consisting of an arbitrary number of zeros,
   followed by one.  Marking makes is possible for the sending HOST to
   synchronize the beginning of the text message with its word
   boundaries.  When the last bit of a message has entered an IMP, the
   hardware interface between the IMP and HOST appends a one followed by
   enough zeros to make the message length a multiple of 16 bits.  These
   appended bits are called padding.  Except for the marking and
   padding, no limitations are placed on the text of a message.  Figure
   2 shows a typical message sent by a 24-bit machine.

DESIGN CONCEPTS

   The computers participating in the network are alike in two important
   respects: each supports research independent of the network, and each
   is under the discipline of a time-sharing system.  These facts
   contributed to the following design philosophy.

   First, because the computers in the network have independent purposes
   it is necessary to preserve decentralized administrative control of
   the various computers.  Since all of the time-sharing supervisors
   possess elaborate and definite accounting and resource allocation



Crocker, et. al.                                                [Page 4]

RFC 33                   New HOST-HOST Protocol         12 February 1970


   mechanisms, we arranged matters so that these mechanisms would
   control the load due to the network in the same way that they control
   locally generated load.

   Second, because the computers are all operated under time-sharing
   disciplines, it seemed desirable to facilitate basic interactive
   mechanisms.

   Third, because this network is used by experienced programmers it was
   imperative to provide the widest latitude in using the network.
   Restrictions concerning character sets, programming languages, etc.
   would not be tolerated and we avoided such restrictions.

   Fourth, again because the network is used by experienced programmers,
   it was felt necessary to leave the design open-ended.  We expect that
   conventions will arise from time to time as experience is gained, but
   we felt constrained not to impose them arbitrarily.

   Fifth, in order to make network participation comfortable, or in some
   cases, feasible, the software interface to the network should require
   minimal surgery on the HOST operating system.

   Finally, we except the assumption stated above that network use
   consists of prolonged conversations instead of one-shot requests.

   These considerations led to the notions of connections, a Network
   Control Program, a control link, control commands, sockets, and
   virtual nets.

   A connection is an extension of a link.  A connection connects two
   processes so that output from one process is input to the other.
   Connections are simplex, so two connections are needed if two
   processes are to converse in both directions.

   Processes within a HOST communicate with the network through a
   Network Control Program (NCP).  In most HOSTs, the NCP will be a part
   of the executive, so that processes will use system calls to
   communicate with it.  The primary function of the NCP is to establish
   connections, break connections, switch connections, and control flow.

   In order to accomplish its tasks, a NCP in one HOST must communicate
   with a NCP in another HOST.  To this end, a particular link between
   each pair of HOSTs has been designated as the control link.  Messages
   received over the control link are always interpreted by the NCP as a
   sequence of one or more control commands.  As an example, one of the
   kinds of control commands is used to assign a link and initiate a





Crocker, et. al.                                                [Page 5]

RFC 33                   New HOST-HOST Protocol         12 February 1970


   connection, while another kind carries notification that a connection
   has been terminated.  A partial sketch of the syntax and semantics of
   control commands is given in the next section.

   A major issue is how to refer to processes in a foreign HOST.  Each
   HOST has some internal naming scheme, but these various schemes often
   are incompatible.  Since it is not practical to impose a common
   internal process naming scheme, an intermediate name space was
   created with a separate portion of the name space given to each HOST.
   It is left to each HOST to map internal process identifiers into its
   name space.

   The elements of the name space are called sockets.  A socket forms
   one end of a connection, and a connection is fully specified by a
   pair of sockets.  A socket is specified by the concatenation of three
   numbers:

      (a) a user number (24 bits)
      (b) a HOST number (8 bits)
      (c) AEN (8 bits)

   A typical socket is illustrated in Figure 3.

   Each HOST is assigned all sockets in the name space which have field
   (b) equal to the HOST's own identification.

   A socket is either a receive socket or a send socket, and is so
   marked by the lower-order bit of the AEN (0 = receive, 1 = send).
   The other seven bits of the AEN simply provide a sizable population
   of sockets for each used number at each HOST.  (AEN stands for
   "another eight-bit number")

   Each user is assigned a 24-bit user number which uniquely identifies
   him throughout the network.  Generally this will be the 8-bit HOST
   number of his home HOST, followed by 16 bits which uniquely identify
   him at that HOST.  Provision can also be made for a user to have a
   user number not keyed to a particular HOST, an arrangement desirable
   for mobile users who might have no home HOST or more than one home
   HOST.  This 24-bit user number is then used in the following manner.
   When a user signs onto a HOST, his user number is looked up.
   Thereafter, each process the user creates is tagged with his user
   number.  When the user signs onto a foreign HOST via the network, his
   same user number is used to tag processes he creates in that HOST.
   The foreign HOST obtains the user number either by consulting a table
   at login time, as the home HOST does, or by noticing the
   identification of the caller.  The effect of propagating the user's
   number is that each user creates his own virtual net consisting of
   processes he has created.  This virtual net may span an arbitrary



Crocker, et. al.                                                [Page 6]

RFC 33                   New HOST-HOST Protocol         12 February 1970


   number of HOSTs.  It will thus be easy for a user to connect his
   processes in arbitrary ways, while still permitting him to connect
   his processes with those in other virtual nets.

   The relationship between sockets and processes is now describable
   (see Figure 4).  For each user number at each HOST, there are 128
   send sockets and 128 receive sockets.  A process may request from the
   local NCP the use of any one of the sockets with the same user
   number; the request is granted if the socket is not otherwise in use.
   The key observation here is that a socket requested by a process
   cannot already be in use unless it is by some other process within
   the same virtual net, and such a process is controlled by the same
   user.

   An unusual aspect of the HOST-HOST protocol is that a process may
   switch its end of a connection from one socket to another.  The new
   socket may be in any virtual net and at any HOST, and the process may
   initiate a switch either at the time the connection is being
   established, or later.  The most general forms of switching entail
   quite complex implementation, and are not germane to the rest of this
   paper, so only a limited form will be explained.  This limited form
   of switching provides only that a process may substitute one socket
   for another while establishing a connection.  The new socket must
   have the same user number and HOST number, and the connection is
   still established to the same process.  This form of switching is
   thus only a way of relabelling a socket, for no charge in the routing
   of messages takes place.  In the next section we document the system
   calls and control commands; in the section after next, we consider
   how login might be implemented.

SYSTEM CALLS AND CONTROL COMMANDS

   Here we sketch the mechanisms of establishing, switching and breaking
   a connection.  As noted above, the NCP interacts with user processes
   via system calls and with other NCPs via control commands.  We
   therefore begin with a partial description of system calls and
   control commands.

   System calls will vary from one operating system to another, so the
   following description is only suggestive.  We assume here that a
   process has several input-output paths which we will call ports.
   Each port may be connected to a sequential I/O device, and while
   connected, transmits information in only one direction.  We further
   assume that the process is blocked (dismissed, slept) while
   transmission proceeds.  The following is the list of system calls:






Crocker, et. al.                                                [Page 7]

RFC 33                   New HOST-HOST Protocol         12 February 1970


            Init      <port>, <AEN 1>, <AEN 2>, <foreign socket>

      where <port> is part of the process issuing the Init
                     _
            <AEN 1>   |
      and             +- are 8-bit AEN's (see Figure 2)
            <AEN 2>   |
                     _|

            The first AEN is used to initiate the connection; the second
            is used while the connection exists.

            <foreign socket> is the 40-bit socket name of the distant
            end of the connection.

            The lower-order bits of <AEN 1> and <AEN 2> must agree, and
            these must be the complement of the lower-order bit of
            <foreign socket>.

            The NCP concatenates <AEN 1> and <AEN 2> each with the user
            number of the process and the HOST number to form 40-bit
            sockets.  It then sends a Request for Connection (RFC)
            control command to the distant NCP.  When the distant NCP
            responds positively, the connection is established and the
            process is unblocked.  If the distant NCP responds
            negatively, the local NCP unblocks the requesting process,
            but informs it that the system call has failed.

            Listen <port>, <AEN 1>

      where <port> and <AEN 1> are as above.  The NCP retains the ports
            and <AEN 1> and blocks the process.  When an RFC control
            command arrives naming the local socket, the process is
            unblocked and notified that a foreign process is calling.

            Accept <AEN 2>

            After a Listen has been satisfied, the process may either
            refuse the call or accept it and switch it to another
            socket.  To accept the call, the process issues the Accept
            system call.  The NCP then sends back an RFC control
            command.

            Close <port>

            After establishing a connection, a process issues a Close to
            break the connection.  The Close is also issued after a
            Listen to refuse a call.



Crocker, et. al.                                                [Page 8]

RFC 33                   New HOST-HOST Protocol         12 February 1970


            Transmit <port>, <addr>

            If <port> is attached to a send socket, <addr> points to a
            message to be sent.  This message is preceded by its length
            in bits.

            If <port> is attached to a receive socket, a message is
            stored at <addr>.  The length of the message is stored
            first.

Control Commands

   A vocabulary of control commands has been defined for communication
   between Network Control Programs.  Each control command consists of
   an 8-bit operation code to indicate its function, followed by some
   parameters.  The number and format of parameters is fixed for each
   operation code.  A sequence of control commands destined for a
   particular HOST can be packed into a single control message.

      RFC   <my socket 1>, <my socket 2>.

            <your socket>, (<link>)

   This command is sent because a process has executed either an Init
   system call or an Accept system call.  A link is assigned by the
   prospective receiver, so it is omitted if <my socket 1> is a send
   socket.

   There is distinct advantage in using the same commands both to
   initiate a connection (Init) and to accept a call (Accept).  If the
   responding command were different from the initiating command, then
   two processes could call each other and become blocked waiting for
   each other to respond.  With this scheme, no deadlock occurs and it
   provides a more compact way to connect a set of processes.

      CLS      <my socket>, <your socket>

   The specified connection is terminated

      CEASE    <link>

   When the receiving process does not consume its input as fast as it
   arrives, the buffer space in the receiving HOST is used to queue the
   waiting messages.  Since only limited space is generally available,
   the receiving HOST may need to inhibit the sending HOST from sending
   any more messages over the offending connection.  When the sending
   HOST receives this command, it may block the process generating the
   messages.



Crocker, et. al.                                                [Page 9]

RFC 33                   New HOST-HOST Protocol         12 February 1970


      RESUME   <link>

   This command is also sent from the receiving HOST to the sending HOST
   and negates a previous CEASE.

LOGGING IN

   We assume that within each HOST there is always a process in
   execution which listens to login requests.  We call this process the
   logger, and it is part of a special virtual net whose user number is
   zero.  The logger is programmed to listen to calls on socket number
   0.  Upon receiving a call, the logger switches it to a higher (even)
   numbered sockets, and returns a call to the socket numbered one less
   than the send socket originally calling.  In this fashion, the logger
   can initiate 127 conversations.

   To illustrate, assume a user whose identification is X'010005' (user
   number 5 at UCLA) signs into UCLA, starts up one of his programs, and
   this program wants to start a process at SRI.  No process except the
   logger is currently willing to listen to our user, so he executes

         Init, <port> = 1, <AEN 1> = 7, <AEN 2> = 7,

               <foreign socket> = 0

   His process is blocked, and the NCP at UCLA sends

         RFC   <my socket 1> = X'0100050107',

               <my socket 2> = X'0100050107',

               <your socket> = X'000000200'

   The logger at SRI is notified when this message is received, because
   it has previously executed

         Listen   <port> = 9, <AEN 1> = 0.

   The logger then executes

         Accept   <AEN 2> = 88.










Crocker, et. al.                                               [Page 10]

RFC 33                   New HOST-HOST Protocol         12 February 1970


   In response to the Accept, the SRI NCP sends

         RFC   <my socket 1> = X'0000000200'

               <my socket 2> = X'0000000258'

               <your socket> = X'0100050107'

               <link> = 37

   where the link has been chosen from the set of available links.  The
   SRI logger than executes

         Init     <port> = 10

               <AEN 1> = 89, <AEN 2> = 89,

               <foreign socket> = X'0100050106'

   which causes the NCP to send

         RFC   <my socket 1> = X'0000000259'

               <my socket 2> = x'0000000259'

               <your socket> = X'0100050106'

   The process at UCLA is unblocked and notified of the successful Init.
   Because SRI logger always initiates a connection to the AEN one less
   than it has just been connected to, the UCLA process then executes

         Listen   <port> = 11

               <AEN 1> = 6

   and when unblocked

         Accept   <AEN 2> = 6

   When these transactions are complete, the UCLA process is doubly
   connected to the logger at SRI.  The logger will then interrogate the
   UCLA process, and if satisfied, create a new process at SRI.  This
   new process will be tagged with user number X'010005', and both
   connections wil be switched to the new process.  In this case,
   switching the connections to the new process corresponds to "passing
   the console down" in many time-sharing systems.





Crocker, et. al.                                               [Page 11]

RFC 33                   New HOST-HOST Protocol         12 February 1970


USER LEVEL SOFTWARE

   At the user level, subroutines which manage data buffer and format
   input designed for other HOSTs are provided.  It is not mandatory
   that the user use such subroutines, since the user has access to the
   network system calls in his monitor.

   In addition to user programming access, it is desirable to have a
   subsystem program at each HOST which makes the network immediately
   accessible from a teletype-like device without special programming.
   Subsystems are commonly used system components such as text editors,
   compilers and interpreters.  An example of a network-related
   subsystem is TELNET, which will allow users at the University of Utah
   to connect to Stanford Research Institute and appear as regular
   terminal users.  It is expected that more sophisticated subsystems
   will be developed in time, but this basic one will render the early
   network immediately useful.

   A user at the University of Utah (UTAH) is sitting at a teletype
   dialed into the University's PDP-10/50 time-sharing system.  He
   wishes to operate the Conversational Algebraic Language (CAL)
   subsystem on the XDS-940 at Stanford Research Institute (SRI) in
   Menlo Park, California.  A typical TELNET dialog is illustrated in
   Figure 5.  The meaning of each line of dialogue is discussed here.

      (i)      The user signs in at UTAH

      (ii)     The PDP-10 run command starts up the TELNET subsystem at
               the user's HOST.

      (111)    The user identifies a break character which causes any
               message following the break to be interpreted locally
               rather than being sent on the foreign HOST.

      (iv)     The TELNET subsystem will make the appropriate system
               calls to establish a pair of connections to the SRI
               logger.  The connections will be established only if SRI
               accepts another foreign user.

   The UTAH user is now in the pre-logged-in state at SRI.  This is
   analogous to the standard teletype user's state after dialing into a
   computer and making a connection but before typing anything.

      (v)      The user signs in to SRI with a standard login command.
               Characters typed on the user's teletype are transmitted
               unaltered through the PDP-10 (user HOST) and on to the
               940 (serving HOST).  The PDP-10 TELNET will have
               automatically switched to full-duplex, character-by-



Crocker, et. al.                                               [Page 12]

RFC 33                   New HOST-HOST Protocol         12 February 1970


               character transmission, since this is required by SRI's
               940.  Full duplex operation is allowed for by the PDP-10,
               though not used by most Digital Equipment Corporations
               subsystems.

      (vi) and (vii) The 940 subsystem, CAL, is started.

   At this point, the user wishes to load a local CAL file into the 940
   CAL subsystem, from the file system on his local PDP-10.

      (viii)   CAL is instructed to establish a connection to UTAH in
               order to receive this file.  "NETWRK" is a predefined 940
               name similar in nature to "PAPER TYPE" or "TELETYPE".

      (ix)     Finally, the user types the break character (#) followed
               by a command to his PDP-10 TELNET program, which sends
               the desired file to SRI from Utah on the connection just
               established for this purpose.  The user's next statement
               is in CAL again.

   The TELNET subsystem coding should be minimal for it is essentially a
   shell program built over the network system calls.  It effectively
   established a shunt in the user HOST between the remote user and a
   distant serving HOST.

   Given the basic system primitives, the TELNET subsystem at the user
   HOST and a manual for the serving HOST, the network can be profitably
   employed by remote users today.

HIGHER LEVEL PROTOCOL

   The network poses special problems where a high degree of interaction
   is required between the user and a particular subsystem in a foreign
   HOST.  These problems arise due to heterogeneous consoles, local
   operating systems overhead, and network transmission delays.  Unless
   we use special strategies it may be difficult or even impossible for
   a distant user to make use of the more sophisticated subsystems
   offered.  While these difficulties are especially severe in the area
   of graphics, problems may arise even for teletype interaction.  For
   example, suppose that a foreign subsystem is designed for teletype
   consoles connected by telephone, and then this subsystem becomes
   available to network users.  This subsystem might have the following
   characteristics.

      1. Except for echoing and correction of mistyping, no action is
         taken until a carriage return is typed.







Crocker, et. al.                                               [Page 13]

RFC 33                   New HOST-HOST Protocol         12 February 1970


      2. All characters except "^", and "<-" and carriage returns are
         echoed as the character is typed.



      3. <- causes deletion of the immediately preceding character, and
         is echoed as that character.



      4. ^ causes all previously typed characters to be ignored.  A
         carriage return and line feed are echoed.



      5. A carriage return is echoed as a carriage return followed by a
         line feed.



   If each character typed is sent in its own message, then the
   characters

      H E L L O <- <- P c.r.

   cause nine messages in each direction.  Furthermore, each character
   is handled by a user level program in the local HOST before being
   sent to the foreign HOST.

   Now it is clear that if this particular example were important, we
   would quickly implement rules 1 to 5 in a local HOST program and send
   only complete lines to the foreign HOST.  If the foreign HOST program
   could not be modified so as to not generate echoes, then the local
   program could not only echo properly, it could also throw away the
   later echoes from the foreign HOST.  However, the problem is not any
   particular interaction scheme; the problem is that we expect many of
   these kinds of schemes to occur.  We have not found any general
   solutions to these problems, but some observations and conjectures
   may lead the way.

   With respect to heterogeneous consoles, we note that although
   consoles are rarely compatible, many are equivalent.  It is probably
   reasonable to treat a model 37 teletype as the equivalent of an IBM
   2741.  Similarly, most storage scopes will form an equivalence class,
   and most refresh display scopes will form another.  Furthermore, a
   hierarchy might emerge with members of one class usable in place of
   those in another, but not vice versa.  We can imagine that any scope
   might be an adequate substitute for a teletype, but hardly the
   reverse.  This observation leads us to wonder if a network-wide
   language for consoles might be possible.  Such a language would
   provide for distinct treatment of different classes of consoles, with
   semantics appropriate to each class.  Each site could then write
   interface programs for its consoles to make them look like network
   standard devices.





Crocker, et. al.                                               [Page 14]

RFC 33                   New HOST-HOST Protocol         12 February 1970


   Another observation is that a user evaluates an interactive system by
   comparing the speed of the system's responses with his own
   expectations.  Sometimes a user feels that he has made only a minor
   request, so the response should be immediate; at other times he feels
   he has made a substantial request, and is therefore willing to wait
   for the response.  Some interactive subsystems are especially
   pleasant to use because a great deal of work has gone into tailoring
   the responses to the user's expectations.  In the network, however, a
   local user level process intervenes between a local console and a
   foreign subsystem, and we may expect the response time for minor
   requests to degrade.  Now it may happen that all of this tailoring of
   the interaction is fairly independent of the portion of the subsystem
   which does the heavy computing or I/O.  In such a case, it may be
   possible to separate a subsystem into two sections.  One section
   would be a "front end" which formats output to the user, accepts his
   input, and controls computationally simple responses such as echoes.
   In the example above, the program to accumulate a line and generate
   echoes would be the front end of some subsystem.  We now take notice
   of the fact that the local HOSTs have substantial computational
   power, but our current designs make use of the local HOST only as a
   data concentrator.  This is somewhat ironic, for the local HOST is
   not only poorly utilized as a data concentrator, it also degrades
   performance because of the delays it introduces.

   These arguments have led us to consider the possibility of a Network
   Interface Language (NIL) which would be a network-wide language for
   writing the front end of interactive subsystems.  This language would
   have the feature that subprograms communicate through network-like
   connections.  The strategy is then to transport the source code for
   the front end of a subsystem to the local HOST, where it would be
   compiled and executed.

   During preliminary discussions we have agreed that NIL should have at
   least the following semantic properties not generally found in other
   languages.

      1. Concurrency.  Because messages arrive asynchronously on
         different connections, and because user input is not
         synchronized with subsystem output, NIL must include semantics
         to accurately model the possible concurrencies.

      2. Program Concatenation.  It is very useful to be able to insert
         a program in between two other programs.  To achieve this, the
         interconnection of programs would be specified at run time and
         would not be implicit in the source code.






Crocker, et. al.                                               [Page 15]

RFC 33                   New HOST-HOST Protocol         12 February 1970


      3. Device substitutability.  It is usual to define languages so
         that one device may be substituted for another.  The
         requirement here is that any device can be modeled by a NIL
         program.  For example, if a network standard display controller
         manipulates tree-structures according to messages sent to it
         then these structures must be easily implementable in NIL.

   NIL has not been fully specified, and reservations have been
   expressed about its usefulness.  These reservations hinge upon our
   conjecture that it is possible to divide an interactive system into a
   transportable front end which satisfies a user's expectations at low
   cost and a more substantial stay-at-home section.  If our conjecture
   is false, then NIL will not be useful; otherwise it seems worth
   pursuing.  Testing of this conjecture and further development of NIL
   will take priority after low level HOST-HOST protocol has stabilized.

HOST/IMP INTERFACING

   The hardware and software interfaces between HOST and IMP is an area
   of particular concern for the HOST organizations.  Considering the
   diversity of HOST computers to which a standard IMP must connect, the
   hardware interface was made bit serial and full-duplex.  Each HOST
   organization implements its half of this very simple interface.

   The software interface is equally simple and consists of messages
   passed back and forth between the IMP and HOST programs.  Special
   error and signal messages are defined as well as messages containing
   normal data.  Messages waiting in queues in either machine are sent
   at the pleasure of the machine in which they reside with no concern
   for the needs of the other computer.

   The effect of the present software interface is the needless
   rebuffering of all messages in the HOST in addition to the buffering
   in the IMP.  The messages have no particular order other than arrival
   times at the IMP.  The Network Control Program at one HOST (e.g.,
   UTAH) needs waiting RFNM's before all other messages.  At another
   site (e.g., SRI), the NCP could benefit by receiving messages for the
   user who is next to be run.

   What is needed is coding representing the specific needs of the HOST
   on both sides of the interface to make intelligent decisions about
   what to transmit next over the channel.  With the present software
   interface, the channel in one direction once committed to a
   particular message is then locked up for up to 80 milliseconds!  This
   approaches one teletype character time and needlessly limits full-
   duplex, character by character, interactions over the net.  At the
   very least, the IMP/HOST protocol should be expended to permit each
   side to assist the other in scheduling messages over the channels.



Crocker, et. al.                                               [Page 16]

RFC 33                   New HOST-HOST Protocol         12 February 1970


CONCLUSIONS

   At this time (February 1970) the initial network of four sites is
   just beginning to be utilized.  The communications system of four
   IMPs and wide band telephone lines have been operational for two
   months.  Programmers at UCLA have signed in as users of the SRI 940.
   More significantly, one of the authors (S. Carr) living in Palo Alto
   uses the Salt Lake PDP-10 on a daily basis by first connecting to
   SRI.  We thus have first hand experience that remote interaction is
   possible and is highly effective.

   Work on the ARPA network has generated new areas of interest.  NIL is
   one example, and interprocess communication is another.  Interprocess
   communication over the network is a subcase of general interprocess
   communication in a multiprogrammed environment.  The mechanism of
   connections seems to be new, and we wonder whether this mechanism is
   useful even when the processes are within the same computer.

REFERENCES

   1     L. ROBERTS
         "The ARPA network"
         Invitational Workshop on Networks of Computers Proceedings
         National Security Agency 1968 p 115 ff

   2.    R M RUTLEDGE et al
         "An interactive network of time-sharing computers"
         Proceedings of the 24th National Conference
         Association for Computing Machinery 1969 p 431 ff

   3.    F E HEART  R E KAHN  S M ORNSTEIN  W R CROWTHER
         D C WALDEN
         "The interface message processors for the ARPA network"
         These Proceedings

LIST OF FIGURES

   Figure 1  Initial network configuration

   Figure 2  A typical message from a 24-bit machine

   Figure 3  A typical socket

   Figure 4  The relationship between sockets and processes

   Figure 5  A typical TELNET dialog.

             Underlined characters are those types by the user.



Crocker, et. al.                                               [Page 17]

RFC 33                   New HOST-HOST Protocol         12 February 1970


                                 SRI
                                _____
                               /     \
                              |  XDS  |
                              |  940  |
                               \_____/
                                  |
                            +----------+
                            |    IMP   |
                            +----------+
                             /   |    \
                            /    |     \
                           /     |      \  +----+    _____
                          /      |       \ | I  |   /     \
       ______     +----+ /       |        \| M  |--|  DEC  |
      /      \    | I  |/        |         | P  |  | PDP-10|
     |   IBM  |---| M  |         |         +----+   \_____/
     | 360/75 |   | P  |\        |
      \______/    +----+ \       |                    UTAH
                          \      |
        UCSB               \     |
                          +----------+
                          |    IMP   |
                          +----------+
                              |
                           ___|___
                          /       \
                         |   XDS   |
                         |(sigma)-7|
                          \_______/

                            UCLA

   Figure 1 Initial network configuration

















Crocker, et. al.                                               [Page 18]

RFC 33                   New HOST-HOST Protocol         12 February 1970


   |<------------ 24bits ----------->|
   |                                 |
   +---------------------------------+
   |                                 |
   |        Leader (32 bits)         |
   |               __________________|
   |              | 100 ---    ----0 |<----16 bits of marking
   +--------------+------------------+
   |                                 |
   |                                 |
   |   Text of messages (96 bits)    |
   |                                 |
   +------------------------+--------+
   | 100-----          ----0|
   +-------^----------------+
           |
           |______16 bits of padding added
                  by the interface

   Figure 2  A typical message from a 24-bit machine



          24                    8          8
   +----------------------+-----------+----------+
   |  User Number         |           |          |
   +----------------------+-----------+----------+
                                |          |___AEN
                                |
                                |___HOST number
   Figure 3 A typical socket



              |<--- connection --->|
   +---------+                      +---------+
   |         |        link          |         |
   | process |--(|--------------|)--| process |
   |         |   ^              ^   |         |
   +---------+   |              |   +---------+
                 |              |
             send socket    receive socket

   Figure 4 The relationship between sockets and processes

         [ This RFC was put into machine readable form for entry ]
          [ into the online RFC archives by Lorrie Shiota 08/00]




Crocker, et. al.                                               [Page 19]