RFC 922


Network Working Group                                      Jeffrey Mogul
Request for Comments: 922                    Computer Science Department
                                                     Stanford University
                                                            October 1984

       BROADCASTING INTERNET DATAGRAMS IN THE PRESENCE OF SUBNETS


Status of this Memo



   We propose simple rules for broadcasting Internet datagrams on local
   networks that support broadcast, for addressing broadcasts, and for
   how gateways should handle them.

   This RFC suggests a proposed protocol for the ARPA-Internet
   community, and requests discussion and suggestions for improvements.
   Distribution of this memo is unlimited.

Acknowledgement



   This proposal here is the result of discussion with several other
   people, especially J. Noel Chiappa and Christopher A. Kent, both of
   whom both pointed me at important references.

1. Introduction



   The use of broadcasts, especially on high-speed local area networks,
   is a good base for many applications.  Since broadcasting is not
   covered in the basic IP specification [12], there is no agreed-upon
   way to do it, and so protocol designers have not made use of it. (The
   issue has been touched upon before, e.g. [6], but has not been the
   subject of a standard.)

   We consider here only the case of unreliable, unsequenced, possibly
   duplicated datagram broadcasts (for a discussion of TCP broadcasting,
   see [10].) Even though unreliable and limited in length, datagram
   broadcasts are quite useful [1].

   We assume that the data link layer of the local network supports
   efficient broadcasting.  Most common local area networks do support
   broadcast; for example, Ethernet [7, 5], ChaosNet [9], token ring
   networks [2], etc.

   We do not assume, however, that broadcasts are reliably delivered.
   (One might consider providing a reliable datagram broadcast protocol
   as a layer above IP.) It is quite expensive to guarantee delivery of
   broadcasts; instead, what we assume is that a host will receive most
   of the broadcasts that are sent.  This is important to avoid
   excessive use of broadcasts; since every host on the network devotes
   at least some effort to every broadcast, they are costly.



Mogul                                                           [Page 1]



RFC 922                                                     October 1984
Broadcasting Internet Datagrams in the Presence of Subnets


   When a datagram is broadcast, it imposes a cost on every host that
   hears it.  Therefore, broadcasting should not be used
   indiscriminately, but rather only when it is the best solution to a
   problem.

2. Terminology



   Because broadcasting depends on the specific data link layer in use
   on a local network, we must discuss it with reference to both
   physical networks and logical networks.

   The terms we will use in referring to physical networks are, from the
   point of view of the host sending or forwarding a broadcast:

   Local Hardware Network

      The physical link to which the host is attached.

   Remote Hardware Network

      A physical network which is separated from the host by at least
      one gateway.

   Collection of Hardware Networks

      A set of hardware networks (transitively) connected by gateways.

   The IP world includes several kinds of logical network.  To avoid
   ambiguity, we will use the following terms:

   Internet

      The DARPA Internet collection of IP networks.

   IP Network

      One or a collection of several hardware networks that have one
      specific IP network number.

   Subnet

      A single member of the collection of hardware networks that
      compose an IP network.  Host addresses on a given subnet share an
      IP network number with hosts on all other subnets of that IP
      network, but the local-address part is divided into subnet-number




Mogul                                                           [Page 2]



RFC 922                                                     October 1984
Broadcasting Internet Datagrams in the Presence of Subnets


      and host-number fields to indicate which subnet a host is on.  We
      do not assume a particular division of the local-address part;
      this could vary from network to network.

   The introduction of a subnet level in the addressing hierarchy is at
   variance with the IP specification [12], but as the use of
   addressable subnets proliferates it is obvious that a broadcasting
   scheme should support subnetting.  For more on subnets, see [8].

   In this paper, the term "host address" refers to the host-on-subnet
   address field of a subnetted IP network, or the host-part field
   otherwise.

   An IP network may consist of a single hardware network or a
   collection of subnets; from the point of view of a host on another IP
   network, it should not matter.

3. Why Broadcast?



   Broadcasts are useful when a host needs to find information without
   knowing exactly what other host can supply it, or when a host wants
   to provide information to a large set of hosts in a timely manner.

   When a host needs information that one or more of its neighbors might
   have, it could have a list of neighbors to ask, or it could poll all
   of its possible neighbors until one responds.  Use of a wired-in list
   creates obvious network management problems (early binding is
   inflexible).  On the other hand, asking all of one's neighbors is
   slow if one must generate plausible host addresses, and try them
   until one works.  On the ARPANET, for example, there are roughly 65
   thousand plausible host numbers.  Most IP implementations have used
   wired-in lists (for example, addresses of "Prime" gateways.)
   Fortunately, broadcasting provides a fast and simple way for a host
   to reach all of its neighbors.

   A host might also use a broadcast to provide all of its neighbors
   with some information; for example, a gateway might announce its
   presence to other gateways.

   One way to view broadcasting is as an imperfect substitute for
   multicasting, the sending of messages to a subset of the hosts on a
   network.  In practice, broadcasts are usually used where multicasts
   are what is wanted; datagrams are broadcast at the hardware level,
   but filtering software in the receiving hosts gives the effect of
   multicasting.

   For more examples of broadcast applications, see [1, 3].


Mogul                                                           [Page 3]



RFC 922                                                     October 1984
Broadcasting Internet Datagrams in the Presence of Subnets


4. Broadcast Classes



   There are several classes of IP broadcasting:

      - Single-destination datagrams broadcast on the local hardware
        net: A datagram is destined for a specific IP host, but the
        sending host broadcasts it at the data link layer, perhaps to
        avoid having to do routing.  Since this is not an IP broadcast,
        the IP layer is not involved, except that a host should discard
        datagram not meant for it without becoming flustered (i.e.,
        printing an error message).

      - Broadcast to all hosts on the local hardware net: A
        distinguished value for the host-number part of the IP address
        denotes broadcast instead of a specific host.  The receiving IP
        layer must be able to recognize this address as well as its own.
        However, it might still be useful to distinguish at higher
        levels between broadcasts and non-broadcasts, especially in
        gateways.  This is the most useful case of broadcast; it allows
        a host to discover gateways without wired-in tables, it is the
        basis for address resolution protocols, and it is also useful
        for accessing such utilities as name servers, time servers,
        etc., without requiring wired-in addresses.

      - Broadcast to all hosts on a remote hardware network: It is
        occasionally useful to send a broadcast to all hosts on a
        non-local network; for example, to find the latest version of a
        hostname database, to bootload a host on a subnet without a
        bootserver, or to monitor the timeservers on the subnet.  This
        case is the same as local-network broadcasts; the datagram is
        routed by normal mechanisms until it reaches a gateway attached
        to the destination hardware network, at which point it is
        broadcast.  This class of broadcasting is also known as
        "directed broadcasting", or quaintly as sending a "letter bomb"
        [1].

      - Broadcast to all hosts on a subnetted IP network (Multi-subnet
        broadcasts): A distinguished value for the subnet-number part of
        the IP address is used to denote "all subnets".  Broadcasts to
        all hosts of a remote subnetted IP network are done just as
        directed broadcasts to a single subnet.

      - Broadcast to the entire Internet: This is probably not useful,
        and almost certainly not desirable.





Mogul                                                           [Page 4]



RFC 922                                                     October 1984
Broadcasting Internet Datagrams in the Presence of Subnets


   For reasons of performance or security, a gateway may choose not to
   forward broadcasts; especially, it may be a good idea to ban
   broadcasts into or out of an autonomous group of networks.

5. Broadcast Methods



   A host's IP receiving layer must be modified to support broadcasting.
   In the absence of broadcasting, a host determines if it is the
   recipient of a datagram by matching the destination address against
   all of its IP addresses.  With broadcasting, a host must compare the
   destination address not only against the host's addresses, but also
   against the possible broadcast addresses for that host.

   The problem of how best to send a broadcast has been extensively
   discussed [1, 3, 4, 13, 14].  Since we assume that the problem has
   already been solved at the data link layer, an IP host wishing to
   send either a local broadcast or a directed broadcast need only
   specify the appropriate destination address and send the datagram as
   usual.  Any sophisticated algorithms need only reside in gateways.

   The problem of broadcasting to all hosts on a subnetted IP network is
   apparently somewhat harder.  However, even in this case it turns out
   that the best known algorithms require no additional complexity in
   non-gateway hosts.  A good broadcast method will meet these
   additional criteria:

      - No modification of the IP datagram format.

      - Reasonable efficiency in terms of the number of excess copies
        generated and the cost of paths chosen.

      - Minimization of gateway modification, in both code and data
        space.

      - High likelihood of delivery.

   The algorithm that appears best is the Reverse Path Forwarding (RPF)
   method [4].  While RPF is suboptimal in cost and reliability, it is
   quite good, and is extremely simple to implement, requiring no
   additional data space in a gateway.









Mogul                                                           [Page 5]



RFC 922                                                     October 1984
Broadcasting Internet Datagrams in the Presence of Subnets


6. Gateways and Broadcasts



   Most of the complexity in supporting broadcasts lies in gateways.  If
   a gateway receives a directed broadcast for a network to which it is
   not connected, it simply forwards it using the usual mechanism.
   Otherwise, it must do some additional work.

   6.1. Local Broadcasts



      When a gateway receives a local broadcast datagram, there are
      several things it might have to do with it.  The situation is
      unambiguous, but without due care it is possible to create
      infinite loops.

      The appropriate action to take on receipt of a broadcast datagram
      depends on several things: the subnet it was received on, the
      destination network, and the addresses of the gateway.

         - The primary rule for avoiding loops is "never broadcast a
           datagram on the hardware network it was received on". It is
           not sufficient simply to avoid repeating datagram that a
           gateway has heard from itself; this still allows loops if
           there are several gateways on a hardware network.

         - If the datagram is received on the hardware network to which
           it is addressed, then it should not be forwarded.  However,
           the gateway should consider itself to be a destination of the
           datagram (for example, it might be a routing table update.)

         - Otherwise, if the datagram is addressed to a hardware network
           to which the gateway is connected, it should be sent as a
           (data link layer) broadcast on that network.  Again, the
           gateway should consider itself a destination of the datagram.

         - Otherwise, the gateway should use its normal routing
           procedure to choose a subsequent gateway, and send the
           datagram along to it.












Mogul                                                           [Page 6]



RFC 922                                                     October 1984
Broadcasting Internet Datagrams in the Presence of Subnets


   6.2. Multi-subnet broadcasts



      When a gateway receives a broadcast meant for all subnets of an IP
      network, it must use the Reverse Path Forwarding algorithm to
      decide what to do.  The method is simple: the gateway should
      forward copies of the datagram along all connected links, if and
      only if the datagram arrived on the link which is part of the best
      route between the gateway and the source of the datagram.
      Otherwise, the datagram should be discarded.

      This algorithm may be improved if some or all of the gateways
      exchange among themselves additional information; this can be done
      transparently from the point of view of other hosts and even other
      gateways.  See [4, 3] for details.

   6.3. Pseudo-Algol Routing Algorithm



      This is a pseudo-Algol description of the routing algorithm a
      gateway should use.  The algorithm is shown in figure 1.  Some
      definitions are:

      RouteLink(host)

         A function taking a host address as a parameter and returning
         the first-hop link from the gateway to the host.

      RouteHost(host)

         As above but returns the first-hop host address.

      ResolveAddress(host)

         Returns the hardware address for an IP host.

      IncomingLink

         The link on which the packet arrived.

      OutgoingLinkSet

         The set of links on which the packet should be sent.

      OutgoingHardwareHost

         The hardware host address to send the packet to.




Mogul                                                           [Page 7]



RFC 922                                                     October 1984
Broadcasting Internet Datagrams in the Presence of Subnets


      Destination.host

         The host-part of the destination address.

      Destination.subnet

         The subnet-part of the destination address.

      Destination.ipnet

         The IP-network-part of the destination address.






































Mogul                                                           [Page 8]



RFC 922                                                     October 1984
Broadcasting Internet Datagrams in the Presence of Subnets

BEGIN
   IF Destination.ipnet IN AllLinks THEN
      BEGIN
         IF IsSubnetted(Destination.ipnet) THEN
            BEGIN
               IF Destination.subnet = BroadcastSubnet THEN
                  BEGIN      /* use Reverse Path Forwarding algorithm */
                     IF IncomingLink = RouteLink(Source) THEN
                        BEGIN IF Destination.host = BroadcastHost THEN
                              OutgoingLinkSet <- AllLinks -
                           IncomingLink;
                           OutgoingHost <- BroadcastHost;
                           Examine packet for possible internal use;
                        END
                     ELSE  /* duplicate from another gateway, discard */
                        Discard;
                  END
               ELSE
                  IF Destination.subnet = IncomingLink.subnet THEN
                     BEGIN           /* forwarding would cause a loop */
                        IF Destination.host = BroadcastHost THEN
                           Examine packet for possible internal use;
                        Discard;
                     END
                  ELSE BEGIN    /* forward to (possibly local) subnet */
                        OutgoingLinkSet <- RouteLink(Destination);
                        OutgoingHost <- RouteHost(Destination);
                     END
            END
         ELSE BEGIN         /* destined for one of our local networks */
               IF Destination.ipnet = IncomingLink.ipnet THEN
                  BEGIN              /* forwarding would cause a loop */
                     IF Destination.host = BroadcastHost THEN
                        Examine packet for possible internal use;
                     Discard;
                  END
               ELSE BEGIN                     /* might be a broadcast */
                     OutgoingLinkSet <- RouteLink(Destination);
                     OutgoingHost <- RouteHost(Destination);
                  END
            END
      END
   ELSE BEGIN                    /* forward to a non-local IP network */
         OutgoingLinkSet <- RouteLink(Destination);
         OutgoingHost <- RouteHost(Destination);
      END
   OutgoingHardwareHost <- ResolveAddress(OutgoingHost);
END

Figure 1: Pseudo-Algol algorithm for routing broadcasts by gateways

Mogul                                                           [Page 9]



RFC 922                                                     October 1984
Broadcasting Internet Datagrams in the Presence of Subnets


7. Broadcast IP Addressing - Conventions



   If different IP implementations are to be compatible, there must be
   convention distinguished number to denote "all hosts" and "all
   subnets".

   Since the local network layer can always map an IP address into data
   link layer address, the choice of an IP "broadcast host number" is
   somewhat arbitrary.  For simplicity, it should be one not likely to
   be assigned to a real host.  The number whose bits are all ones has
   this property; this assignment was first proposed in [6].  In the few
   cases where a host has been assigned an address with a host-number
   part of all ones, it does not seem onerous to require renumbering.

   The "all subnets" number is also all ones; this means that a host
   wishing to broadcast to all hosts on a remote IP network need not
   know how the destination address is divided up into subnet and host
   fields, or if it is even divided at all.  For example, 36.255.255.255
   may denote all the hosts on a single hardware network, or all the
   hosts on a subnetted IP network with 1 byte of subnet field and 2
   bytes of host field, or any other possible division.

   The address 255.255.255.255 denotes a broadcast on a local hardware
   network that must not be forwarded.  This address may be used, for
   example, by hosts that do not know their network number and are
   asking some server for it.

   Thus, a host on net 36, for example, may:

      - broadcast to all of its immediate neighbors by using
        255.255.255.255

      - broadcast to all of net 36 by using 36.255.255.255

   without knowing if the net is subnetted; if it is not, then both
   addresses have the same effect. A robust application might try the
   former address, and if no response is received, then try the latter.
   See [1] for a discussion of such "expanding ring search" techniques.

   If the use of "all ones" in a field of an IP address means
   "broadcast", using "all zeros" could be viewed as meaning
   "unspecified".  There is probably no reason for such addresses to
   appear anywhere but as the source address of an ICMP Information
   Request datagram.  However, as a notational convention, we refer to
   networks (as opposed to hosts) by using addresses with zero fields.
   For example, 36.0.0.0 means "network number 36" while 36.255.255.255
   means "all hosts on network number 36".


Mogul                                                          [Page 10]



RFC 922                                                     October 1984
Broadcasting Internet Datagrams in the Presence of Subnets


   7.1. ARP Servers and Broadcasts



      The Address Resolution Protocol (ARP) described in [11] can, if
      incorrectly implemented, cause problems when broadcasts are used
      on a network where not all hosts share an understanding of what a
      broadcast address is.  The temptation exists to modify the ARP
      server so that it provides the mapping between an IP broadcast
      address and the hardware broadcast address.

      This temptation must be resisted.  An ARP server should never
      respond to a request whose target is a broadcast address.  Such a
      request can only come from a host that does not recognize the
      broadcast address as such, and so honoring it would almost
      certainly lead to a forwarding loop.  If there are N such hosts on
      the physical network that do not recognize this address as a
      broadcast, then a datagram sent with a Time-To-Live of T could
      potentially give rise to T**N spurious re-broadcasts.

8. References



   1.   David Reeves Boggs.  Internet Broadcasting.  Ph.D. Th., Stanford
        University, January 1982.

   2.   D.D. Clark, K.T. Pogran, and D.P. Reed.  "An Introduction to
        Local Area Networks".  Proc. IEEE 66, 11, pp1497-1516,
        November 1978.

   3.   Yogan Kantilal Dalal.  Broadcast Protocols in Packet Switched
        Computer Networks.  Ph.D. Th., Stanford University, April 1977.

   4.   Yogan K. Dalal and Robert M. Metcalfe.  "Reverse Path Forwarding
        of Broadcast Packets".  Comm. ACM 21, 12, pp1040-1048,
        December 1978.

   5.   The Ethernet, A Local Area Network: Data Link Layer and Physical
        Layer Specifications.  Version 1.0, Digital Equipment
        Corporation, Intel, Xerox, September 1980.

   6.   Robert Gurwitz and Robert Hinden.  IP - Local Area Network
        Addressing Issues.  IEN-212, BBN, September 1982.

   7.   R.M. Metcalfe and D.R. Boggs.  "Ethernet: Distributed Packet
        Switching for Local Computer Networks".  Comm. ACM 19, 7,
        pp395-404, July 1976.  Also CSL-75-7, Xerox Palo Alto Research
        Center, reprinted in CSL-80-2.




Mogul                                                          [Page 11]



RFC 922                                                     October 1984
Broadcasting Internet Datagrams in the Presence of Subnets


   8.   Jeffrey Mogul.  Internet Subnets.  RFC-917, Stanford University,
        October 1984.

   9.   David A. Moon.  Chaosnet.  A.I. Memo 628, Massachusetts
        Institute of Technology Artificial Intelligence Laboratory,
        June 1981.



   10.  William W. Plummer.  Internet Broadcast Protocols.  IEN-10, BBN,
        March 1977.



   11.  David Plummer.  An Ethernet Address Resolution Protocol.
        RFC-826, Symbolics, September 1982.



   12.  Jon Postel.  Internet Protocol.  RFC-791, ISI, September 1981.



   13.  David W. Wall.  Mechanisms for Broadcast and Selective
        Broadcast.  Ph.D. Th., Stanford University, June 1980.



   14.  David W. Wall and Susan S. Owicki.  Center-based Broadcasting.
        Computer Systems Lab Technical Report TR189, Stanford
        University, June 1980.






























Mogul                                                          [Page 12]