All files are for educational and/or historic purposes only. [back to library]

Date: Wed, 12 Nov 86 21:59:15 pst
From: wally%[email protected] (Wally Linstruth) (tty00)

        To: tcpgroup
        Regarding: IP addressing

             The  intent  of  this paper is to  document  the  background 
        behind the current IP address assignments which I have offered to 
        coordinate.   The proposed scheme has been reviewed by Phil Karn, 
        Bdale Garbee and (verbally with) Mike Chepponis, all of whom have 
        encouraged that it be used.

             Phil's  code  does  NOT  currently  support  the  subnetwork 
        aspects of the scheme but will do so in the future.   There is no 
        real  reason  for  any national coordination of  these  addresses 
        until  actual  networks or at  least  geographically  coordinated 
        groups of experimenters are formed.

             I  have offered to issue and keep track of SUBNET  addresses 
        and  their  "owners"  who are  presumably  responsible  *NETWORK* 
        implementors and managers.

             The  basic premise behind the proposed plan is that  amateur 
        radio  networks will be politically defined.   The plan is  based 
        upon  the  presumption  that current voice networks  serve  as  a 
        proper  analog by which to predict general characteristics of the 
        as yet unconstructed digital networks.   Political entities  will 
        build networks; funded, controlled, maintained and used primarily 
        by their own members and guests.

             Each  of these separately managed networks should be  viewed 
        as  a  subnetwork  of AMPRNET (with the  idea  being  to  somehow 
        rationally  partition the AMPRNET address space).  
        Each  subnetwork within AMPRNET will maintain routing tables  for 
        its  own  constituents.   Each will provide its own hosts  (TACs, 
        Gateways, i.e. the mechanism by which users with simple terminals 
        and AX25 level 2 boxes will access network resources),  switches, 
        rules  (network  administration),  security  measures  and  quite 
        possibly its own link level protocols.

             The  natural  limitations on span of control  will  probably 
        limit  the  service  area of each of  these  networks.   This  is 
        another factor leading to the partitioning of the AMPRNET address 
        space with respect to separate subnetworks.

             This  partitioning  of  the  address space  will  allow  for 
        much  simplified  routing tables in each  host.   Internetworking 
        gateways will connect these independently controlled subnetworks. 
        Each  gateway will maintain routing tables only for  local  hosts 
        and for gateways to other networks.  Hosts and  relay switches on 
        a   given  subnet  will  need  to  maintain  routing  information 
        regarding  only  members  of that subnet and  gateways  to  other 
        networks.   The  required routing tables should prove to be  very 
        manageable  and make any kind of geographically  based  hueristic 
        addressing schemes such as ZIP codes, area codes etc. moot.


             I  would  also  like to propose that we  coordinate  logical 
        network  names and their corresponding addresses based  on  these 
        political   network  subdivisions.    The  concept  of  a  naming 
        convention  which maps directly into an IP address is purely  for 
        the  convenience  of  network developers and  is  not  considered 
        necessary.  There is,  however, some good reasoning behind making 
        network and host names hierarchical and meaningful to end  users.  
        It  will  considerably aid in bootstrapping the initial  networks 
        and in being comprehensible to the non-network folks who will  be 
        the  primary  users  of these networks.   The  naming  convention 
        proposed   is  of  the   form   [email protected][.AMPRNET.RES].  
        WESTNET,  SBARCnet  (Santa  Barbara ARC) and  GFRN-net  represent 
        three  hypothetical  networks  with which this  writer  could  be 
        involved, perhaps as a provider of gateway and/or host services.

             Each  of  these  subnetwork entities could have  a  distinct 
        address  and  perhaps several internally  administered  host/user 

             [NOTE:  Throughout this paper,  Host or Host/User represents 
             any  host  or any user running IP protocols that has  direct 
             network  access.  Also,  for the purposes of  the  following 
             example,   WA6JPR  is  not  a  network  address,  rather  it 
             represents  a user-id on a local host.   It is the  writer's 
             opinion that the majority of packet users for the forseeable 
             future  will  be using simple TNCs connected  to  hosts  via 
             AX.25 level 2 protocols.]

             WA6JPR  may  be "a user" on hosts on more than  one  network 
        such  that  a station in Washington D.C.,logged onto  an  AMPRNET 
        host,    may    send    internet    traffic    successfully    to 
        [email protected]  (this traffic would be routed to  Westnet 
        via various AMPRNET gateways and subnetwork level relays and then 
        to  a  Santa  Barbara  host known internally  by  Westnet  to  be 
        reachable  via  the  W6AMT-2  switch).   Traffic  could  also  be 
        directed  to  [email protected]   (presuming  that  the  Santa  Barbara 
        Amateur  Radio Club maintains a message server host gatewayed  to 
        the AMPRNET catenet).

             Based  upon  the  presumption  of  the   AMPRNET/SUBNET/HOST 
        hierarchy,  it  would  seem  that we could easily decide  how  to 
        allocate  the 24 bit IP address field  such  that 
        there  are bits allocated for a sufficient number of individually 
        managed  subnetworks  while leaving  a  correspondingly  adequate 
        number  of  assignable bits for the internal addressing needs  of 
        each individual subnetwork.

             Accordingly,   the  following  is  proposed  as  an  initial 
        addressing  scheme and methodology for address assignment.   [Bit 
        numbering is per RFC-960 Pg.2]


        Bit  8  to  be 0 for USA stations and  1  for  non-USA  stations.  
        [Note.   This  is  not  meant  to imply a  geographic  basis  for 
        assignments.   It  is  meant to provide a very  quick  means  for 
        segregating FCC controlled participants from non-FCC stations.]

        Bits  9 - 18 to represent politically separate subnetworks within 
        AMPRNET.   These  bits  are to be assigned in an  inverse  binary 
        sequence   (see   example   below)  beginning  with   the   *MOST 
        SIGNIFICANT* bit first.

        Bits 19 - 23 to be unassigned and reserved for future  allocation 
        as  network addresses,  to network administrations for internally 
        assigned  host  and/or user addresses,  to a combination  of  the 
        above or to a completely new intermediate class of addresses.

        Bits  24  - 31  to be used within  politically  separate  AMPRNET 
        subnetworks for individual hosts,  switches, workstations etc. as 
        determined  by  local  network  administration.    It  would   be 
        recommended  that these bits be assigned in binary sequence  with 
        the *LEAST SIGNIFICANT* bits being assigned first.

             The  resulting  network  addresses would be as follows:

        || SUBNET----+
        || |         |
        || |         | HOST--+
        || |         | |     |
        44:0...127:000:0...255------- 32,768 addresses assignable
                    |           +- 1,015,808 addresses reserved
        44:0...127:032:0...255------- 32,768 addresses assignable
                    |           +- 1,015,808 addresses reserved
        44:0...127:064:0...255------- 32,768 addresses assignable
                    |           +- 1,015,808 addresses reserved
        44:0...127:096:0...255------- 32,768 addresses assignable
                    |           +- 1,015,808 addresses reserved
        44:0...127:128:0...255------- 32,768 addresses assignable
                    |           +- 1,015,808 addresses reserved
        44:0...127:160:0...255------- 32,768 addresses assignable
                    |           +- 1,015,808 addresses reserved
        44:0...127:192:0...255------- 32,768 addresses assignable


                    |           +- 1,015,808 addresses reserved
        44:0...127:224:0...255------- 32,768 addresses assignable
                    |           +- 1,015,808 addresses reserved

            |                   +- 8,388,608 addresses assignable (non USA)

             The  above  allocation and assignment scheme allows  network 
        (subnet)  and  intranet  (host/user) addresses  to  begin  to  be 
        immediately assigned to experimenters while retaining the largest 
        possible  contiguous  block of unassigned bits whose  assignments 
        can  be  defined  in  the future with  little  or  no  impact  on 
        previously   allocated   addresses.    The  USER  @  HOSTNAME   . 
        SUBNET/ADMINISTRATION  naming scheme represents a  human-friendly 
        network  naming  convention  which  maps  easily  into  numerical 
        network  addresses.   I  believe that the above  approach  is  in 
        general  conformance with the requirements of RFC-950,  "Internet 
        Standard Subnetting Procedure."

             The  numbering scheme as initially proposed allows for up to 
        1024  AMPRNET  subnetworks  of up to 256 hosts in the  USA  while 
        retaining  five  bits  for  future  expansion.    That's  262,144 
        individual AMPRNET addressable entities.   If the proposed method 
        of  address  assignment is followed and we run out  of  Host/User 
        addresses before we run out of network addresses,  we can  simply 
        pick  up  the  least  significant reserved bit  and  assign  more 
        Host/User addresses.   Conversely,  if network addresses are more 
        popular  we  could easily expand by taking the  most  significant 
        reserved  bit and allocating it for network  addressing.

        If it should become clear that every user on a network needs  his 
        or her own IP address, each network could allocate user blocks in 
        256  user  increments from the least significant  reserved  bits.  
        Possible  combinations  are  1024 networks each with up  to  8192 
        individually  addressable units or 2048 networks each  with  4096 
        hosts/users (8,388,608 individually addressable entities).

             The  writer presumes that 8 million plus addresses ought  to 
        last the US amateur population for some time to come. All we need 
        to  do to avoid painting ourselves in a corner is to assign  them 
        in a logical sequence rather than randomly.


             The  following table serves as an example of the  "high  bit 
        first"  network  address  assignment table and  some  actual  and 
        requested initial networking assignments.

             "this"      ;special case
             KARNnet      ;network admin: KA9Q
             BDALEnet      ;network admin: N3EUA
             DCnet1      ;network admin: WB6RQN
             SOCALnet1      ;network admin: WB5EKU
             DCnet2      ;network admin: WB6RQN
             SOCALnet2      ;network admin: WA6JPR
             PITTNET      ;network admin: N3CVL
             "all"      ;special case