INITIAL CONNECTION PROTOCOL--REVIEWED

INTRODUCTION

------------

At the Network meeting preceding the SJCC '71, an
"ICP Committee" was established. It's purpose was to get
"something" working fast with minimum modifications to the

current ICP so as to minimize complaints. (This seems like

a good definition for almost everything!) Consequently,

those who had objections to the current ICP were interviewed

and a compromise was reached in the form of RFC #165. The

ICP committee didn't have a chance to think about an alter-

native because of the above mentioned constraints. In this

note we attempted a simple version of an ICP assuming that

we can add commands to Host-Host protocol. We hope that this

will be useful in the design of the next version of the

Host-Host protocol.

ICP COMMANDS

------------

To establish a regular connection one party can issue

an INIT (NCP sends RTS or STR commands), then the other

party can accept the request for connection by responding

with an INIT or refusing it with a CLOSE. We think that

[Page 1]

---

a similar, simple mechanism is desirable for the ICP.

Furthermore, the ICP should allow for simplex as well as

duplex connections from user to server.

The following commands are necessary for simplex

connections:

ISC - Initiate Simplex Connection

ASC - Accept Simplex Connection

RSC - Refuse Simplex Connection

The notation for parameters is similar to that

of RFC #165:

L - Server socket name, in one special case the
server is "logger".

U - User socket.

S - Socket assigned by server for the connection
with user.

X - Is the byte size if U is odd and is the link
number if U is even.

     -                           -
     X - Is the complement of X (X is the link number
         if U is odd and byte size if U is even.

To initiate a simplex connection the user's NCP

issues:

ISC, L, U, X

To refuse this connection the server's NCP issues:

RSC, L, U

[Page 2]

---

To accept this connection the server's NCP issues:

                         -
           ASC, L, U, S, X

Similarly, for duplex connections, we have:

IDC, L, U1, X1, U2, X2

RDC, L, U1, U2

                           -           -
           ADC, L, U1, S1, X1, U2, S2, X2

where (U1,U2), (S1,S2), (U1,S1) and (U2,S2) are pairs of

opposite gender.

After the server accepts the connection(s), it (they)

goes immediately to a "connected state", and the appropriate

ALL command(s) must be sent.

ADVANTAGES

----------

The main advantage to this approach is that it mini-

mizes the dialog between user and server. The server socket

L is used only as an address, not a socket to connect to,

therefore eliminating the need to establish a connection to

L, choose a byte size, send an ALL command, send and receive

data on it and CLOSE it. Race conditions as mentioned in

RFC #143 do not arise. Socket L is the server and should

be in a "Listening for ICP" state when an ISC or IDC is

received. If socket L is not in that state, the serving

NCP should refuse to ICP request. The serving NCP should

_not_ queue ICP requests.

In the current ICP, when the user choses socket U, he

has to reserve sockets U+2 and U+3. In the above described

approach no restrictions exist for U1 and U2 (except that

they are of opposite gender); the same is true for S1 and

S2.

We think that duplex commands are necessary since both

connections are to be connected to the same server process.

Their separation by using two ISCs, will add complications

of correlating the two ISCs with the same process. Also,

if two ISCs are used, the first might be accepted and the

second refused. This leads to uncertainty on the user's

part. This condition cannot occur with the duplex commands.

[Page 3]

---

MINIMUM MODIFICATION TO CURRENT ICP

-----------------------------------

The minimum change we can think of to make the current

ICP look similar to the above is to add one NCP level com-

mand -- accept:

ACC, L, U, S

The exchange between NCPs in the notation of RFC #165

is now:

<where the original uses a script lowercase "L" we use "l">

<where the original uses subscripts we use {} so that
   A-subscript-B is printed A{B} >

     Server NCP                   User NCP
     ----------                   --------

     Listen for connection on L   RST,U,L,l{A}

     ACC,L,U,S                    S is passed by NCP to the
                                  user and connection from
                                  U to L is closed.

     STR,S+1,U+2,B{s}             STR,U+3,S,B{u}

     RTS,S,U+3,l{B}               RTS,U+2,S+1,l{c}

     Wait for connection          Wait for connection

     ALL,l{B},m{B},b{B}           ALL,l{c},m{c},b{c}

An alternative way to the ACC command is a CLS command

with an additional parameter (32 bits long). If parameter

is zero the request for connection by the user is refused;

if the parameter is non-zero, the request is accepted and

socket S is the value of the parameter.

All suggested changes improve the ICP dialog both from

the aesthetic and efficiency points of view. We lean strongly,

however, to the first, more major ICP modification.

[Page 4]

---

A COMMENT ABOUT CLS COMMAND

---------------------------

It seems appropriate to mention here for the purpose

of the next version of the Host-Host protocol that the

CLS command has more than one function. We think that the

CLS command should be reserved to close connections in the

"connected state" only (i.e., "open" connections). Two

additional commands can be used for "refusing" and "reject-

ing" requests for connections:

REJ<mysocket><yoursocket> -- when a request
for connection is rejected unconditionally.

REF<mysocket><yoursocket><reason> -- when a
request for connection is refused temporarily
because the NCP could not handle it. For
example: no process LISTENed to it and it
was timed-out, or NCP tables are full in which
case the user process may try again. The
reason for refusing is indicated in the
parameter "reason".

[ This RFC was put into machine readable form for entry ]

[ into the online RFC archives by BBN Corp. under the   ]
[ direction of Alex McKenzie.                   12/96   ]

[Page 5]

---