RFC 1163
This document is obsolete. Please refer to RFC 1267.






Network Working Group                                        K. Lougheed
Request for Comments: 1163                                 cisco Systems
Obsoletes: RFC 1105                                           Y. Rekhter
                                   T.J. Watson Research Center, IBM Corp
                                                               June 1990


                    A Border Gateway Protocol (BGP)

Status of this Memo



   This RFC, together with its companion RFC-1164, "Application of the
   Border Gateway Protocol in the Internet", define a Proposed Standard
   for an inter-autonomous system routing protocol for the Internet.

   This protocol, like any other at this initial stage, may undergo
   modifications before reaching full Internet Standard status as a
   result of deployment experience.  Implementers are encouraged to
   track the progress of this or any protocol as it moves through the
   standardization process, and to report their own experience with the
   protocol.

   This protocol is being considered by the Interconnectivity Working
   Group (IWG) of the Internet Engineering Task Force (IETF).
   Information about the progress of BGP can be monitored and/or
   reported on the IWG mailing list (IWG@nri.reston.va.us).

   Please refer to the latest edition of the "IAB Official Protocol
   Standards" RFC for current information on the state and status of
   standard Internet protocols.

   Distribution of this memo is unlimited.

Table of Contents



      1.  Acknowledgements......................................    2
      2.  Introduction..........................................    2
      3.  Summary of Operation..................................    4
      4.  Message Formats.......................................    5
      4.1 Message Header Format.................................    5
      4.2 OPEN Message Format...................................    6
      4.3 UPDATE Message Format.................................    8
      4.4 KEEPALIVE Message Format..............................   10
      4.5 NOTIFICATION Message Format...........................   10
      5.  Path Attributes.......................................   12
      6.  BGP Error Handling....................................   14
      6.1 Message Header error handling.........................   14
      6.2 OPEN message error handling...........................   15



Lougheed & Rekhter                                              [Page 1]

RFC 1163                          BGP                          June 1990


      6.3 UPDATE message error handling.........................   16
      6.4 NOTIFICATION message error handling...................   17
      6.5 Hold Timer Expired error handling.....................   17
      6.6 Finite State Machine error handling...................   18
      6.7 Cease.................................................   18
      7.  BGP Version Negotiation...............................   18
      8.  BGP Finite State machine..............................   18
      9.  UPDATE Message Handling...............................   22
      10. Detection of Inter-AS Policy Contradictions...........   23
      Appendix 1.  BGP FSM State Transitions and Actions........   25
      Appendix 2.  Comparison with RFC 1105.....................   28
      Appendix 3.  TCP options that may be used with BGP........   28
      References................................................   29
      Security Considerations...................................   29
      Authors' Addresses........................................   29

1.  Acknowledgements



   We would like to express our thanks to Guy Almes (Rice University),
   Len Bosack (cisco Systems), Jeffrey C. Honig (Cornell Theory Center)
   and all members of the Interconnectivity Working Group of the
   Internet Engineering Task Force, chaired by Guy Almes, for their
   contributions to this document.

   We would also like to thank Bob Hinden, Director for Routing of the
   Internet Engineering Steering Group, and the team of reviewers he
   assembled to review earlier versions of this document.  This team,
   consisting of Deborah Estrin, Milo Medin, John Moy, Radia Perlman,
   Martha Steenstrup, Mike St. Johns, and Paul Tsuchiya, acted with a
   strong combination of toughness, professionalism, and courtesy.

2.  Introduction



   The Border Gateway Protocol (BGP) is an inter-Autonomous System
   routing protocol.  It is built on experience gained with EGP as
   defined in RFC 904 [1] and EGP usage in the NSFNET Backbone as
   described in RFC 1092 [2] and RFC 1093 [3].

   The primary function of a BGP speaking system is to exchange network
   reachability information with other BGP systems.  This network
   reachability information includes information on the full path of
   Autonomous Systems (ASs) that traffic must transit to reach these
   networks.  This information is sufficient to construct a graph of AS
   connectivity from which routing loops may be pruned and some policy
   decisions at the AS level may be enforced.

   To characterize the set of policy decisions that can be enforced
   using BGP, one must focus on the rule that an AS advertize to its



Lougheed & Rekhter                                              [Page 2]

RFC 1163                          BGP                          June 1990


   neighbor ASs only those routes that it itself uses.  This rule
   reflects the "hop-by-hop" routing paradigm generally used throughout
   the current Internet.  Note that some policies cannot be supported by
   the "hop-by-hop" routing paradigm and thus require techniques such as
   source routing to enforce.  For example, BGP does not enable one AS
   to send traffic to a neighbor AS intending that that traffic take a
   different route from that taken by traffic originating in the
   neighbor AS.  On the other hand, BGP can support any policy
   conforming to the "hop-by-hop" routing paradigm.  Since the current
   Internet uses only the "hop-by-hop" routing paradigm and since BGP
   can support any policy that conforms to that paradigm, BGP is highly
   applicable as an inter-AS routing protocol for the current Internet.

   A more complete discussion of what policies can and cannot be
   enforced with BGP is outside the scope of this document (but refer to
   the companion document discussing BGP usage [5]).

   BGP runs over a reliable transport protocol.  This eliminates the
   need to implement explicit update fragmentation, retransmission,
   acknowledgement, and sequencing.  Any authentication scheme used by
   the transport protocol may be used in addition to BGP's own
   authentication mechanisms.  The error notification mechanism used in
   BGP assumes that the transport protocol supports a "graceful" close,
   i.e., that all outstanding data will be delivered before the
   connection is closed.

   BGP uses TCP [4] as its transport protocol.  TCP meets BGP's
   transport requirements and is present in virtually all commercial
   routers and hosts.  In the following descriptions the phrase
   "transport protocol connection" can be understood to refer to a TCP
   connection.  BGP uses TCP port 179 for establishing its connections.

   This memo uses the term `Autonomous System' (AS) throughout.  The
   classic definition of an Autonomous System is a set of routers under
   a single technical administration, using an interior gateway protocol
   and common metrics to route packets within the AS, and using an
   exterior gateway protocol to route packets to other ASs.  Since this
   classic definition was developed, it has become common for a single
   AS to use several interior gateway protocols and sometimes several
   sets of metrics within an AS.  The use of the term Autonomous System
   here stresses the fact that, even when multiple IGPs and metrics are
   used, the administration of an AS appears to other ASs to have a
   single coherent interior routing plan and presents a consistent
   picture of what networks are reachable through it.  From the
   standpoint of exterior routing, an AS can be viewed as monolithic:
   reachability to networks directly connected to the AS must be
   equivalent from all border gateways of the AS.




Lougheed & Rekhter                                              [Page 3]

RFC 1163                          BGP                          June 1990


   The planned use of BGP in the Internet environment, including such
   issues as topology, the interaction between BGP and IGPs, and the
   enforcement of routing policy rules is presented in a companion
   document [5].  This document is the first of a series of documents
   planned to explore various aspects of BGP application.

3.  Summary of Operation



   Two systems form a transport protocol connection between one another.
   They exchange messages to open and confirm the connection parameters.
   The initial data flow is the entire BGP routing table.  Incremental
   updates are sent as the routing tables change.  BGP does not require
   periodic refresh of the entire BGP routing table.  Therefore, a BGP
   speaker must retain the current version of the entire BGP routing
   tables of all of its peers for the duration of the connection.
   KeepAlive messages are sent periodically to ensure the liveness of
   the connection.  Notification messages are sent in response to errors
   or special conditions.  If a connection encounters an error
   condition, a notification message is sent and the connection is
   closed.

   The hosts executing the Border Gateway Protocol need not be routers.
   A non-routing host could exchange routing information with routers
   via EGP or even an interior routing protocol.  That non-routing host
   could then use BGP to exchange routing information with a border
   router in another Autonomous System.  The implications and
   applications of this architecture are for further study.

   If a particular AS has multiple BGP speakers and is providing transit
   service for other ASs, then care must be taken to ensure a consistent
   view of routing within the AS.  A consistent view of the interior
   routes of the AS is provided by the interior routing protocol.  A
   consistent view of the routes exterior to the AS can be provided by
   having all BGP speakers within the AS maintain direct BGP connections
   with each other.  Using a common set of policies, the BGP speakers
   arrive at an agreement as to which border routers will serve as
   exit/entry points for particular networks outside the AS.  This
   information is communicated to the AS's internal routers, possibly
   via the interior routing protocol.  Care must be taken to ensure that
   the interior routers have all been updated with transit information
   before the BGP speakers announce to other ASs that transit service is
   being provided.

   Connections between BGP speakers of different ASs are referred to as
   "external" links.  BGP connections between BGP speakers within the
   same AS are referred to as "internal" links.





Lougheed & Rekhter                                              [Page 4]

RFC 1163                          BGP                          June 1990


4.  Message Formats



   This section describes message formats used by BGP.

   Messages are sent over a reliable transport protocol connection.  A
   message is processed only after it is entirely received.  The maximum
   message size is 4096 octets.  All implementations are required to
   support this maximum message size.  The smallest message that may be
   sent consists of a BGP header without a data portion, or 19 octets.

4.1 Message Header Format



   Each message has a fixed-size header.  There may or may not be a data
   portion following the header, depending on the message type.  The
   layout of these fields is shown below:

    0                   1                   2                   3
    0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |                                                               |
   +                                                               +
   |                                                               |
   +                                                               +
   |                           Marker                              |
   +                                                               +
   |                                                               |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |          Length               |      Type     |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

   Marker:

      This 16-octet field contains a value that the receiver of the
      message can predict.  If the Type of the message is OPEN, or if
      the Authentication Code used in the OPEN message of the connection
      is zero, then the Marker must be all ones.  Otherwise, the value
      of the marker can be predicted by some a computation specified as
      part of the authentication mechanism used.  The Marker can be used
      to detect loss of synchronization between a pair of BGP peers, and
      to authenticate incoming BGP messages.

   Length:

      This 2-octet unsigned integer indicates the total length of the
      message, including the header, in octets.  Thus, e.g., it allows
      one to locate in the transport-level stream the (Marker field of
      the) next message.  The value of the Length field must always be
      at least 19 and no greater than 4096, and may be further



Lougheed & Rekhter                                              [Page 5]

RFC 1163                          BGP                          June 1990


      constrained, depending on the message type.  No "padding" of extra
      data after the message is allowed, so the Length field must have
      the smallest value required given the rest of the message.

   Type:

      This 1-octet unsigned integer indicates the type code of the
      message.  The following type codes are defined:

                           1 - OPEN
                           2 - UPDATE
                           3 - NOTIFICATION
                           4 - KEEPALIVE

4.2 OPEN Message Format



   After a transport protocol connection is established, the first
   message sent by each side is an OPEN message.  If the OPEN message is
   acceptable, a KEEPALIVE message confirming the OPEN is sent back.
   Once the OPEN is confirmed, UPDATE, KEEPALIVE, and NOTIFICATION
   messages may be exchanged.

   In addition to the fixed-size BGP header, the OPEN message contains
   the following fields:

     0                   1                   2                   3
     0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
    +-+-+-+-+-+-+-+-+
    |    Version    |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    |     My Autonomous System      |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    |           Hold Time           |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    |  Auth. Code   |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    |                                                               |
    |                       Authentication Data                     |
    |                                                               |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

   Version:

      This 1-octet unsigned integer indicates the protocol version
      number of the message.  The current BGP version number is 2.






Lougheed & Rekhter                                              [Page 6]

RFC 1163                          BGP                          June 1990


   My Autonomous System:

      This 2-octet unsigned integer indicates the Autonomous System
      number of the sender.

   Hold Time:

      This 2-octet unsigned integer indicates the maximum number of
      seconds that may elapse between the receipt of successive
      KEEPALIVE and/or UPDATE and/or NOTIFICATION messages.

   Authentication Code:

      This 1-octet unsigned integer indicates the authentication
      mechanism being used.  Whenever an authentication mechanism is
      specified for use within BGP, three things must be included in the
      specification:
         - the value of the Authentication Code which indicates use of
         the mechanism,
         - the form and meaning of the Authentication Data, and
         - the algorithm for computing values of Marker fields.
      Only one authentication mechanism is specified as part of this
      memo:
         - its Authentication Code is zero,
         - its Authentication Data must be empty (of zero length), and
         - the Marker fields of all messages must be all ones.
      The semantics of non-zero Authentication Codes lies outside the
      scope of this memo.

      Note that a separate authentication mechanism may be used in
      establishing the transport level connection.

   Authentication Data:

      The form and meaning of this field is a variable-length field
      depend on the Authentication Code.  If the value of Authentication
      Code field is zero, the Authentication Data field must have zero
      length.  The semantics of the non-zero length Authentication Data
      field is outside the scope of this memo.

      Note that the length of the Authentication Data field can be
      determined from the message Length field by the formula:

         Message Length = 25 + Authentication Data Length

      The minimum length of the OPEN message is 25 octets (including
      message header).




Lougheed & Rekhter                                              [Page 7]

RFC 1163                          BGP                          June 1990


4.3 UPDATE Message Format



   UPDATE messages are used to transfer routing information between BGP
   peers.  The information in the UPDATE packet can be used to construct
   a graph describing the relationships of the various Autonomous
   Systems.  By applying rules to be discussed, routing information
   loops and some other anomalies may be detected and removed from
   inter-AS routing.

   In addition to the fixed-size BGP header, the UPDATE message contains
   the following fields (note that all fields may have arbitrary
   alignment):

     0                   1                   2                   3
     0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    |  Total Path Attributes Length |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    |                                                               |
    /                      Path Attributes                          /
    /                                                               /
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    |                       Network 1                               |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    /                                                               /
    /                                                               /
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    |                       Network n                               |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

   Total Path Attribute Length:

      This 2-octet unsigned integer indicates the total length of the
      Path Attributes field in octets.  Its value must allow the (non-
      negative integer) number of Network fields to be determined as
      specified below.

   Path Attributes:

      A variable length sequence of path attributes is present in every
      UPDATE.  Each path attribute is a triple <attribute type,
      attribute length, attribute value> of variable length.

      Attribute Type is a two-octet field that consists of the Attribute
      Flags octet followed by the Attribute Type Code octet.






Lougheed & Rekhter                                              [Page 8]

RFC 1163                          BGP                          June 1990


       0                   1
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |  Attr. Flags  |Attr. Type Code|
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

      The high-order bit (bit 0) of the Attribute Flags octet is the
      Optional bit.  It defines whether the attribute is optional (if
      set to 1) or well-known (if set to 0).

      The second high-order bit (bit 1) of the Attribute Flags octet is
      the Transitive bit.  It defines whether an optional attribute is
      transitive (if set to 1) or non-transitive (if set to 0).  For
      well-known attributes, the Transitive bit must be set to 1.  (See
      Section 5 for a discussion of transitive attributes.)

      The third high-order bit (bit 2) of the Attribute Flags octet is
      the Partial bit.  It defines whether the information contained in
      the optional transitive attribute is partial (if set to 1) or
      complete (if set to 0).  For well-known attributes and for
      optional non-transitive attributes the Partial bit must be set to
      0.

      The fourth high-order bit (bit 3) of the Attribute Flags octet is
      the Extended Length bit.  It defines whether the Attribute Length
      is one octet (if set to 0) or two octets (if set to 1).  Extended
      Length may be used only if the length of the attribute value is
      greater than 255 octets.

      The lower-order four bits of the Attribute Flags octet are unused.
      They must be zero (and should be ignored when received).

      The Attribute Type Code octet contains the Attribute Type Code.
      Currently defined Attribute Type Codes are discussed in Section 5.

      If the Extended Length bit of the Attribute Flags octet is set to
      0, the third octet of the Path Attribute contains the length of
      the attribute data in octets.

      If the Extended Length bit of the Attribute Flags octet is set to
      1, then the third and the fourth octets of the path attribute
      contain the length of the attribute data in octets.

      The remaining octets of the Path Attribute represent the attribute
      value and are interpreted according to the Attribute Flags and the
      Attribute Type Code.

      The meaning and handling of Path Attributes is discussed in



Lougheed & Rekhter                                              [Page 9]

RFC 1163                          BGP                          June 1990


      Section 5.

   Network:

      Each 4-octet Internet network number indicates one network whose
      Inter-Autonomous System routing is described by the Path
      Attributes.  Subnets and host addresses are specifically not
      allowed.  The total number of Network fields in the UPDATE message
      can be determined by the formula:

         Message Length = 19 + Total Path Attribute Length + 4 * #Nets

      The message Length field of the message header and the Path
      Attributes Length field of the UPDATE message must be such that
      the formula results in a non-negative integer number of Network
      fields.

   The minimum length of the UPDATE message is 37 octets (including
   message header).

4.4 KEEPALIVE Message Format



   BGP does not use any transport protocol-based keep-alive mechanism to
   determine if peers are reachable.  Instead, KEEPALIVE messages are
   exchanged between peers often enough as not to cause the hold time
   (as advertised in the OPEN message) to expire.  A reasonable maximum
   time between KEEPALIVE messages would be one third of the Hold Time
   interval.

   KEEPALIVE message consists of only message header and has a length of
   19 octets.

4.5 NOTIFICATION Message Format



   A NOTIFICATION message is sent when an error condition is detected.
   The BGP connection is closed immediately after sending it.

   In addition to the fixed-size BGP header, the NOTIFICATION message
   contains the following fields:

     0                   1                   2                   3
     0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    | Error code    | Error subcode |           Data                |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+                               +
    |                                                               |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+




Lougheed & Rekhter                                             [Page 10]

RFC 1163                          BGP                          June 1990


   Error Code:

      This 1-octet unsigned integer indicates the type of NOTIFICATION.
      The following Error Codes have been defined:

           Error Code       Symbolic Name               Reference

             1         Message Header Error             Section 6.1
             2         OPEN Message Error               Section 6.2
             3         UPDATE Message Error             Section 6.3
             4         Hold Timer Expired               Section 6.5
             5         Finite State Machine Error       Section 6.6
             6         Cease                            Section 6.7

   Error subcode:

      This 1-octet unsigned integer provides more specific information
      about the nature of the reported error.  Each Error Code may have
      one or more Error Subcodes associated with it.  If no appropriate
      Error Subcode is defined, then a zero (Unspecific) value is used
      for the Error Subcode field.

      Message Header Error subcodes:

                      1  - Connection Not Synchronized.
                      2  - Bad Message Length.
                      3  - Bad Message Type.

      OPEN Message Error subcodes:

                      1  - Unsupported Version Number.
                      2  - Bad Peer AS.
                      3  - Unsupported Authentication Code.
                      4  - Authentication Failure.

      UPDATE Message Error subcodes:

                      1 - Malformed Attribute List.
                      2 - Unrecognized Well-known Attribute.
                      3 - Missing Well-known Attribute.
                      4 - Attribute Flags Error.
                      5 - Attribute Length Error.
                      6 - Invalid ORIGIN Attribute
                      7 - AS Routing Loop.
                      8 - Invalid NEXT_HOP Attribute.
                      9 - Optional Attribute Error.
                     10 - Invalid Network Field.




Lougheed & Rekhter                                             [Page 11]

RFC 1163                          BGP                          June 1990


   Data:

      This variable-length field is used to diagnose the reason for the
      NOTIFICATION.  The contents of the Data field depend upon the
      Error Code and Error Subcode.  See Section 6 below for more
      details.

      Note that the length of the Data field can be determined from the
      message Length field by the formula:

         Message Length = 21 + Data Length

   The minimum length of the NOTIFICATION message is 21 octets
   (including message header).

5.  Path Attributes



   This section discusses the path attributes of the UPDATE message.

   Path attributes fall into four separate categories:

         1. Well-known mandatory.
         2. Well-known discretionary.
         3. Optional transitive.
         4. Optional non-transitive.

   Well-known attributes must be recognized by all BGP implementations.
   Some of these attributes are mandatory and must be included in every
   UPDATE message.  Others are discretionary and may or may not be sent
   in a particular UPDATE message.  Which well-known attributes are
   mandatory or discretionary is noted in the table below.

   All well-known attributes must be passed along (after proper
   updating, if necessary) to other BGP peers.

   In addition to well-known attributes, each path may contain one or
   more optional attributes.  It is not required or expected that all
   BGP implementations support all optional attributes.  The handling of
   an unrecognized optional attribute is determined by the setting of
   the Transitive bit in the attribute flags octet.  Unrecognized
   transitive optional attributes should be accepted and passed along to
   other BGP peers.  If a path with unrecognized transitive optional
   attribute is accepted and passed along to other BGP peers, the
   Partial bit in the Attribute Flags octet is set to 1.  If a path with
   recognized transitive optional attribute is accepted and passed along
   to other BGP peers and the Partial bit in the Attribute Flags octet
   is set to 1 by some previous AS, it is not set back to 0 by the
   current AS.  Unrecognized non-transitive optional attributes should



Lougheed & Rekhter                                             [Page 12]

RFC 1163                          BGP                          June 1990


   be quietly ignored and not passed along to other BGP peers.

   New transitive optional attributes may be attached to the path by the
   originator or by any other AS in the path.  If they are not attached
   by the originator, the Partial bit in the Attribute Flags octet is
   set to 1.  The rules for attaching new non-transitive optional
   attributes will depend on the nature of the specific attribute.  The
   documentation of each new non-transitive optional attribute will be
   expected to include such rules.  (The description of the INTER-AS
   METRIC attribute gives an example.)  All optional attributes (both
   transitive and non-transitive) may be updated (if appropriate) by ASs
   in the path.

   The order of attributes within the Path Attributes field of a
   particular UPDATE message is irrelevant.

   The same attribute cannot appear more than once within the Path
   Attributes field of a particular UPDATE message.

   Following table specifies attribute type code, attribute length, and
   attribute category for path attributes defined in this document:

   Attribute Name     Type Code    Length     Attribute category
      ORIGIN              1          1        well-known, mandatory
      AS_PATH             2       variable    well-known, mandatory
      NEXT_HOP            3          4        well-known, mandatory
      UNREACHABLE         4          0        well-known, discretionary
      INTER-AS METRIC     5          2        optional, non-transitive

   ORIGIN:

      The ORIGIN path attribute defines the origin of the path
      information.  The data octet can assume the following values:

         Value    Meaning
           0       IGP - network(s) are interior to the originating AS
           1       EGP - network(s) learned via EGP
           2       INCOMPLETE - network(s) learned by some other means

   AS_PATH:

      The AS_PATH attribute enumerates the ASs that must be traversed to
      reach the networks listed in the UPDATE message.  Since an AS
      identifier is 2 octets, the length of an AS_PATH attribute is
      twice the number of ASs in the path.  Rules for constructing an
      AS_PATH attribute are discussed in Section 9.





Lougheed & Rekhter                                             [Page 13]

RFC 1163                          BGP                          June 1990


   NEXT_HOP:

      The NEXT_HOP path attribute defines the IP address of the border
      router that should be used as the next hop to the networks listed
      in the UPDATE message.  This border router must belong to the same
      AS as the BGP peer that advertises it.

   UNREACHABLE:

      The UNREACHABLE attribute is used to notify a BGP peer that some
      of the previously advertised routes have become unreachable.

   INTER-AS METRIC:

      The INTER-AS METRIC attribute may be used on external (inter-AS)
      links to discriminate between multiple exit or entry points to the
      same neighboring AS.  The value of the INTER-AS METRIC attribute
      is a 2-octet unsigned number which is called a metric.  All other
      factors being equal, the exit or entry point with lower metric
      should be preferred.  If received over external links, the INTER-
      AS METRIC attribute may be propagated over internal links to other
      BGP speaker within the same AS.  The INTER-AS METRIC attribute is
      never propagated to other BGP speakers in neighboring AS's.

6.  BGP Error Handling.



   This section describes actions to be taken when errors are detected
   while processing BGP messages.

   When any of the conditions described here are detected, a
   NOTIFICATION message with the indicated Error Code, Error Subcode,
   and Data fields is sent, and the BGP connection is closed.  If no
   Error Subcode is specified, then a zero should be used.

   The phrase "the BGP connection is closed" means that the transport
   protocol connection has been closed and that all resources for that
   BGP connection have been deallocated.  Routing table entries
   associated with the remote peer are marked as invalid.  The fact that
   the routes have become invalid is passed to other BGP peers before
   the routes are deleted from the system.

   Unless specified explicitly, the Data field of the NOTIFICATION
   message that is sent to indicate an error is empty.

6.1 Message Header error handling.



   All errors detected while processing the Message Header are indicated
   by sending the NOTIFICATION message with Error Code Message Header



Lougheed & Rekhter                                             [Page 14]

RFC 1163                          BGP                          June 1990


   Error.  The Error Subcode elaborates on the specific nature of the
   error.

   The expected value of the Marker field of the message header is all
   ones if the message type is OPEN.  The expected value of the Marker
   field for all other types of BGP messages determined based on the
   Authentication Code in the BGP OPEN message and the actual
   authentication mechanism (if the Authentication Code in the BGP OPEN
   message is non-zero). If the Marker field of the message header is
   not the expected one, then a synchronization error has occurred and
   the Error Subcode is set to Connection Not Synchronized.

   If the Length field of the message header is less than 19 or greater
   than 4096, or if the Length field of an OPEN message is less  than
   the minimum length of the OPEN message, or if the Length field of an
   UPDATE message is less than the minimum length of the UPDATE message,
   or if the Length field of a KEEPALIVE message is not equal to 19, or
   if the Length field of a NOTIFICATION message is less than the
   minimum length of the NOTIFICATION message, then the Error Subcode is
   set to Bad Message Length.  The Data field contains the erroneous
   Length field.

   If the Type field of the message header is not recognized, then the
   Error Subcode is set to Bad Message Type.  The Data field contains
   the erroneous Type field.

6.2 OPEN message error handling.



   All errors detected while processing the OPEN message are indicated
   by sending the NOTIFICATION message with Error Code OPEN Message
   Error.  The Error Subcode elaborates on the specific nature of the
   error.

   If the version number contained in the Version field of the received
   OPEN message is not supported, then the Error Subcode is set to
   Unsupported Version Number.  The Data field is a 2-octet unsigned
   integer, which indicates the largest locally supported version number
   less than the version the remote BGP peer bid (as indicated in the
   received OPEN message).

   If the Autonomous System field of the OPEN message is unacceptable,
   then the Error Subcode is set to Bad Peer AS.  The determination of
   acceptable Autonomous System numbers is outside the scope of this
   protocol.

   If the Authentication Code of the OPEN message is not recognized,
   then the Error Subcode is set to Unsupported Authentication Code.




Lougheed & Rekhter                                             [Page 15]

RFC 1163                          BGP                          June 1990


   If the Authentication Code is zero, then the Authentication Data must
   be of zero length.  Otherwise, the Error Subcode is set to
   Authentication Failure.

   If the Authentication Code is non-zero, then the corresponding
   authentication procedure is invoked.  If the authentication procedure
   (based on Authentication Code and Authentication Data) fails, then
   the Error Subcode is set to Authentication Failure.

6.3 UPDATE message error handling.



   All errors detected while processing the UPDATE message are indicated
   by sending the NOTIFICATION message with Error Code UPDATE Message
   Error.  The error subcode elaborates on the specific nature of the
   error.

   Error checking of an UPDATE message begins by examining the path
   attributes.  If the Total Attribute Length is too large (i.e., if
   Total Attribute Length + 21 exceeds the message Length), or if the
   (non-negative integer) Number of Network fields cannot be computed as
   in Section 4.3, then the Error Subcode is set to Malformed Attribute
   List.

   If any recognized attribute has Attribute Flags that conflict with
   the Attribute Type Code, then the Error Subcode is set to Attribute
   Flags Error.  The Data field contains the erroneous attribute (type,
   length and value).

   If any recognized attribute has Attribute Length that conflicts with
   the expected length (based on the attribute type code), then the
   Error Subcode is set to Attribute Length Error.  The Data field
   contains the erroneous attribute (type, length and value).

   If any of the mandatory well-known attributes are not present, then
   the Error Subcode is set to Missing Well-known Attribute.  The Data
   field contains the Attribute Type Code of the missing well-known
   attribute.

   If any of the mandatory well-known attributes are not recognized,
   then the Error Subcode is set to Unrecognized Well-known Attribute.
   The Data field contains the unrecognized attribute (type, length and
   value).

   If the ORIGIN attribute has an undefined value, then the Error
   Subcode is set to Invalid Origin Attribute.  The Data field contains
   the unrecognized attribute (type, length and value).

   If the NEXT_HOP attribute field is syntactically incorrect, then the



Lougheed & Rekhter                                             [Page 16]

RFC 1163                          BGP                          June 1990


   Error Subcode is set to Invalid NEXT_HOP Attribute.  The Data field
   contains the incorrect attribute (type, length and value).  Syntactic
   correctness means that the NEXT_HOP attribute represents a valid IP
   host address.

   The AS route specified by the AS_PATH attribute is checked for AS
   loops.  AS loop detection is done by scanning the full AS route (as
   specified in the AS_PATH attribute) and checking that each AS occurs
   at most once.  If a loop is detected, then the Error Subcode is set
   to AS Routing Loop.  The Data field contains the incorrect attribute
   (type, length and value).

   If an optional attribute is recognized, then the value of this
   attribute is checked.  If an error is detected, the attribute is
   discarded, and the Error Subcode is set to Optional Attribute Error.
   The Data field contains the attribute (type, length and value).

   If any attribute appears more than once in the UPDATE message, then
   the Error Subcode is set to Malformed Attribute List.

   Each Network field in the UPDATE message is checked for syntactic
   validity.  If the Network field is syntactically incorrect, or
   contains a subnet or a host address, then the Error Subcode is set to
   Invalid Network Field.

6.4 NOTIFICATION message error handling.



   If a peer sends a NOTIFICATION message, and there is an error in that
   message, there is unfortunately no means of reporting this error via
   a subsequent NOTIFICATION message.  Any such error, such as an
   unrecognized Error Code or Error Subcode, should be noticed, logged
   locally, and brought to the attention of the administration of the
   peer.  The means to do this, however, lies outside the scope of this
   document.

6.5 Hold Timer Expired error handling.



   If a system does not receive successive KEEPALIVE and/or UPDATE
   and/or NOTIFICATION messages within the period specified in the Hold
   Time field of the OPEN message, then the NOTIFICATION message with
   Hold Timer Expired Error Code should be sent and the BGP connection
   closed.









Lougheed & Rekhter                                             [Page 17]

RFC 1163                          BGP                          June 1990


6.6 Finite State Machine error handling.



   Any error detected by the BGP Finite State Machine (e.g., receipt of
   an unexpected event) is indicated by sending the NOTIFICATION message
   with Error Code Finite State Machine Error.

6.7 Cease.



   In absence of any fatal errors (that are indicated in this section),
   a BGP peer may choose at any given time to close its BGP connection
   by sending the NOTIFICATION message with Error Code Cease.  However,
   the Cease NOTIFICATION message should not be used when a fatal error
   indicated by this section does exist.

7.  BGP Version Negotiation.



   BGP speakers may negotiate the version of the protocol by making
   multiple attempts to open a BGP connection, starting with the highest
   version number each supports.  If an open attempt fails with an Error
   Code OPEN Message Error, and an Error Subcode Unsupported Version
   Number, then the BGP speaker has available the version number it
   tried, the version number its peer tried, the version number passed
   by its peer in the NOTIFICATION message, and the version numbers that
   it supports.  If the two peers do support one or more common
   versions, then this will allow them to rapidly determine the highest
   common version. In order to support BGP version negotiation, future
   versions of BGP must retain the format of the OPEN and NOTIFICATION
   messages.

8.  BGP Finite State machine.



   This section specifies BGP operation in terms of a Finite State
   Machine (FSM).  Following is a brief summary and overview of BGP
   operations by state as determined by this FSM.  A condensed version
   of the BGP FSM is found in Appendix 1.

   Initially BGP is in the Idle state.

      Idle state:

         In this state BGP refuses all incoming BGP connections.  No
         resources are allocated to the BGP neighbor.  In response to
         the Start event (initiated by either system or operator) the
         local system initializes all BGP resources, starts the
         ConnectRetry timer, initiates a transport connection to other
         BGP peer, while listening for connection that may be initiated
         by the remote BGP peer, and changes its state to Connect.




Lougheed & Rekhter                                             [Page 18]

RFC 1163                          BGP                          June 1990


         The exact value of the ConnectRetry timer is a local matter,
         but should be sufficiently large to allow TCP initialization.

         Any other event received in the Idle state is ignored.

      Connect state:

         In this state BGP is waiting for the transport protocol
         connection to be completed.

         If the transport protocol connection succeeds, the local system
         clears the ConnectRetry timer, completes initialization, sends
         an OPEN message to its peer, and changes its state to OpenSent.

         If the transport protocol connect fails (e.g., retransmission
         timeout), the local system restarts the ConnectRetry timer,
         continues to listen for a connection that may be initiated by
         the remote BGP peer, and changes its state to Active state.

         In response to the ConnectRetry timer expired event, the local
         system restarts the ConnectRetry timer, initiates a transport
         connection to other BGP peer, continues to listen for a
         connection that may be initiated by the remote BGP peer, and
         stays in the Connect state.

         Start event is ignored in the Active state.

         In response to any other event (initiated by either system or
         operator), the local system releases all BGP resources
         associated with this connection and changes its state to Idle.

      Active state:

         In this state BGP is trying to acquire a BGP neighbor by
         initiating a transport protocol connection.

         If the transport protocol connection succeeds, the local system
         clears the ConnectRetry timer, completes initialization, sends
         an OPEN message to its peer, sets its hold timer to a large
         value, and changes its state to OpenSent.

         In response to the ConnectRetry timer expired event, the local
         system restarts the ConnectRetry timer, initiates a transport
         connection to other BGP peer, continues to listen for a
         connection that may be be initiated by the remote BGP peer, and
         changes its state to Connect.

         If the local system detects that a remote peer is trying to



Lougheed & Rekhter                                             [Page 19]

RFC 1163                          BGP                          June 1990


         establish BGP connection to it, and the IP address of the
         remote peer is not an expected one, the local system restarts
         the ConnectRetry timer, rejects the attempted connection,
         continues to listen for a connection that may be initiated by
         the remote BGP peer, and stays in the Active state.

         Start event is ignored in the Active state.

         In response to any other event (initiated by either system or
         operator), the local system releases all BGP resources
         associated with this connection and changes its state to Idle.

      OpenSent state:

         In this state BGP waits for an OPEN message from its peer.
         When an OPEN message is received, all fields are checked for
         correctness.  If the BGP message header checking or OPEN
         message checking detects an error (see Section 6), the local
         system sends a NOTIFICATION message and changes its state to
         Idle.

         If there are no errors in the OPEN message, BGP sends a
         KEEPALIVE message and sets a KeepAlive timer.  The hold timer,
         which was originally set to an arbitrary large value (see
         above), is replaced with the value indicated in the OPEN
         message.  If the value of the Autonomous System field is the
         same as our own , then the connection is "internal" connection;
         otherwise, it is "external".  (This will effect UPDATE
         processing as described below.)  Finally, the state is changed
         to OpenConfirm.

         If a disconnect notification is received from the underlying
         transport protocol, the local system closes the BGP connection,
         restarts the ConnectRetry timer, while continue listening for
         connection that may be initiated by the remote BGP peer, and
         goes into the Active state.

         If the hold time expires, the local system sends NOTIFICATION
         message with error code Hold Timer Expired and changes its
         state to Idle.

         In response to the Stop event (initiated by either system or
         operator) the local system sends NOTIFICATION message with
         Error Code Cease and changes its state to Idle.

         Start event is ignored in the OpenSent state.

         In response to any other event the local system sends



Lougheed & Rekhter                                             [Page 20]

RFC 1163                          BGP                          June 1990


         NOTIFICATION message with Error Code Finite State Machine Error
         and changes its state to Idle.

         Whenever BGP changes its state from OpenSent to Idle, it closes
         the BGP (and transport-level) connection and releases all
         resources associated with that connection.

      OpenConfirm state:

         In this state BGP waits for a KEEPALIVE or NOTIFICATION
         message.

         If the local system receives a KEEPALIVE message, it changes
         its state to Established.

         If the hold timer expires before a KEEPALIVE message is
         received, the local system sends NOTIFICATION message with
         error code Hold Timer expired and changes its state to Idle.

         If the local system receives a NOTIFICATION message, it changes
         its state to Idle.

         If the KeepAlive timer expires, the local system sends a
         KEEPALIVE message and restarts its KeepAlive timer.

         If a disconnect notification is received from the underlying
         transport protocol, the local system changes its state to Idle.

         In response to the Stop event (initiated by either system or
         operator) the local system sends NOTIFICATION message with
         Error Code Cease and changes its state to Idle.

         Start event is ignored in the OpenConfirm state.

         In response to any other event the local system sends
         NOTIFICATION message with Error Code Finite State Machine Error
         and changes its state to Idle.

         Whenever BGP changes its state from OpenConfirm to Idle, it
         closes the BGP (and transport-level) connection and releases
         all resources associated with that connection.

      Established state:

         In the Established state BGP can exchange UPDATE, NOTIFICATION,
         and KEEPALIVE messages with its peer.

         If the local system receives an UPDATE or KEEPALIVE message, it



Lougheed & Rekhter                                             [Page 21]

RFC 1163                          BGP                          June 1990


         restarts its Holdtime timer.

         If the local system receives a NOTIFICATION message, it changes
         its state to Idle.

         If the local system receives an UPDATE message and the UPDATE
         message error handling procedure (see Section 6.3) detects an
         error, the local system sends a NOTIFICATION message and
         changes its state to Idle.

         If a disconnect notification is received from the underlying
         transport protocol, the local system  changes its state to
         Idle.

         If the Holdtime timer expires, the local system sends a
         NOTIFICATION message with Error Code Hold Timer Expired and
         changes its state to Idle.

         If the KeepAlive timer expires, the local system sends a
         KEEPALIVE message and restarts its KeepAlive timer.

         Each time the local system sends a KEEPALIVE or UPDATE message,
         it restarts its KeepAlive timer.

         In response to the Stop event (initiated by either system or
         operator), the local system sends a NOTIFICATION message with
         Error Code Cease and changes its state to Idle.

         Start event is ignored in the Established state.

         In response to any other event, the local system sends
         NOTIFICATION message with Error Code Finite State Machine Error
         and changes its state to Idle.

         Whenever BGP changes its state from Established to Idle, it
         closes the BGP (and transport-level) connection, releases all
         resources associated with that connection, and deletes all
         routes derived from that connection.

9.  UPDATE Message Handling



   An UPDATE message may be received only in the Established state.
   When an UPDATE message is received, each field is checked for
   validity as specified in Section 6.3.

   If an optional non-transitive attribute is unrecognized, it is
   quietly ignored.  If an optional transitive attribute is
   unrecognized, the Partial bit (the third high-order bit) in the



Lougheed & Rekhter                                             [Page 22]

RFC 1163                          BGP                          June 1990


   attribute flags octet is set to 1, and the attribute is retained for
   propagation to other BGP speakers.

   If an optional attribute is recognized, and has a valid value, then,
   depending on the type of the optional attribute, it is processed
   locally, retained, and updated, if necessary, for possible
   propagation to other BGP speakers.

   If the network and the path attributes associated with a route to
   that network are correct, then the route is compared with other
   routes to the same network.  If the new route is better than the
   current one, then it is propagated via an UPDATE message to adjacent
   BGP speakers as follows:

   - If a route in the UPDATE was received over an internal link, it is
     not propagated over any other internal link.  This restriction is
     due to the fact that all BGP speakers within a single AS form a
     completely connected graph (see above).

   - If the UPDATE message is propagated over an external link, then the
     local AS number is prepended to the AS_PATH attribute, and the
     NEXT_HOP attribute is updated with an IP address of the router that
     should be used as a next hop to the network.  If the UPDATE message
     is propagated over an internal link, then the AS_PATH attribute is
     passed unmodified and the NEXT_HOP attribute is replaced with the
     sender's own IP address.

   Generally speaking, the rules for comparing routes among several
   alternatives are outside the scope of this document.  There are two
   exceptions:

   - If the local AS appears in the AS path of the new route being
     considered, then that new route cannot be viewed as better than any
     other route.  If such a route were ever used, a routing loop would
     result.

   - In order to achieve successful distributed operation, only routes
     with a likelihood of stability can be chosen.  Thus, an AS must
     avoid using unstable routes, and it must not make rapid spontaneous
     changes to its choice of route.  Quantifying the terms "unstable"
     and "rapid" in the previous sentence will require experience, but
     the principle is clear.

10. Detection of Inter-AS Policy Contradictions



   Since BGP requires no central authority for coordinating routing
   policies among ASs, and since routing policies are not exchanged via
   the protocol itself, it is possible for a group of ASs to have a set



Lougheed & Rekhter                                             [Page 23]

RFC 1163                          BGP                          June 1990


   of routing policies that cannot simultaneously be satisfied.  This
   may cause an indefinite oscillation of the routes in this group of
   ASs.

   To help detect such a situation, all BGP speakers must observe the
   following rule.  If a route to a destination that is currently used
   by the local system is determined to be unreachable (e.g., as a
   result of receiving an UPDATE message for this route with the
   UNREACHABLE attribute), then, before switching to another route, this
   local system must advertize this route as unreachable to all the BGP
   neighbors to which it previously advertized this route.

   This rule will allow other ASs to distinguish between two different
   situations:

   - The local system has chosen to use a new route because the old
     route become unreachable.

   - The local system has chosen to use a new route because it preferred
     it over the old route.  The old route is still viable.

   In the former case, an UPDATE message with the UNREACHABLE attribute
   will be received for the old route.  In the latter case it will not.

   In some cases, this may allow a BGP speaker to detect the fact that
   its policies, taken together with the policies of some other AS,
   cannot simultaneously be satisfied.  For example, consider the
   following situation involving AS A and its neighbor AS B.  B
   advertises a route with a path of the form <B,...>, where A is not
   present in the path.  A then decides to use this path, and advertises
   <A,B,...> to all its neighbors.  B later advertises <B,...,A,...>
   back to A, without ever declaring its previous path <B,...> to be
   unreachable.  Evidently, A prefers routes via B and B prefers routes
   via A.  The combined policies of A and B, taken together, cannot be
   satisfied.  Such an event should be noticed, logged locally, and
   brought to the attention of AS A's administration.  The means to do
   this, however, lies outside the scope of this document.  Also outside
   the document is a more complete procedure for detecting such
   contradictions of policy.

   While the above rules provide a mechanism to detect a set of routing
   policies that cannot be satisfied simultaneously, the protocol itself
   does not provide any mechanism for suppressing the route oscillation
   that may result from these unsatisfiable policies.  The reason for
   doing this is that routing policies are viewed as external to the
   protocol and as determined by the local AS administrator.





Lougheed & Rekhter                                             [Page 24]

RFC 1163                          BGP                          June 1990


Appendix 1.  BGP FSM State Transitions and Actions.

   This Appendix discusses the transitions between states in the BGP FSM
   in response to BGP events.  The following is the list of these states
   and events.

    BGP States:

             1 - Idle
             2 - Connect
             3 - Active
             4 - OpenSent
             5 - OpenConfirm
             6 - Established


    BGP Events:

             1 - BGP Start
             2 - BGP Stop
             3 - BGP Transport connection open
             4 - BGP Transport connection closed
             5 - BGP Transport connection open failed
             6 - BGP Transport fatal error
             7 - ConnectRetry timer expired
             8 - Holdtime timer expired
             9 - KeepAlive timer expired
            10 - Receive OPEN message
            11 - Receive KEEPALIVE message
            12 - Receive UPDATE messages
            13 - Receive NOTIFICATION message

   The following table describes the state transitions of the BGP FSM
   and the actions triggered by these transitions.

    Event                Actions               Message Sent   Next State
    --------------------------------------------------------------------
    Idle (1)
     1            Initialize resources            none             2
                  Start ConnectRetry timer
                  Initiate a transport connection
     others               none                    none             1









Lougheed & Rekhter                                             [Page 25]

RFC 1163                          BGP                          June 1990


    Connect(2)
     1                    none                    none             2
     3            Complete initialization         OPEN             4
                  Clear ConnectRetry timer
     5            Restart ConnectRetry timer      none             3
     7            Restart ConnectRetry timer      none             2
                  Initiate a transport connection
     others       Release resources               none             1

    Active (3)
     1                    none                    none             3
     3            Complete initialization         OPEN             4
                  Clear ConnectRetry timer
     5            Close connection                                 3
                  Restart ConnectRetry timer
     7            Restart ConnectRetry timer      none             2
                  Initiate a transport connection
     others       Release resources               none             1

    OpenSent(4)
     1                    none                    none             4
     4            Close transport connection      none             3
                  Restart ConnectRetry timer
     6            Release resources               none             1
    10            Process OPEN is OK            KEEPALIVE          5
                  Process OPEN failed           NOTIFICATION       1
    others        Close transport connection    NOTIFICATION       1
                  Release resources

    OpenConfirm (5)
     1                   none                     none             5
     4            Release resources               none             1
     6            Release resources               none             1
     9            Restart KeepAlive timer       KEEPALIVE          5
    11            Complete initialization         none             6
                  Restart Holdtime timer
    13            Close transport connection                       1
                  Release resources
    others        Close transport connection    NOTIFICATION       1
                  Release resources











Lougheed & Rekhter                                             [Page 26]

RFC 1163                          BGP                          June 1990


    Established (6)
     1                   none                     none             6
     4            Release resources               none             1
     6            Release resources               none             1
     9            Restart KeepAlive timer       KEEPALIVE          6
    11            Restart Holdtime timer        KEEPALIVE          6
    12            Process UPDATE is OK          UPDATE             6
                  Process UPDATE failed         NOTIFICATION       1
    13            Close transport connection                       1
                  Release resources
    others        Close transport connection    NOTIFICATION       1
                  Release resources
   ---------------------------------------------------------------------

   The following is a condensed version of the above state transition
   table.

Events| Idle | Active | Connect | OpenSent | OpenConfirm | Estab
      | (1)  |   (2)  |  (3)    |    (4)   |     (5)     |   (6)
      |--------------------------------------------------------------
 1    |  2   |    2   |   3     |     4    |      5      |    6
      |      |        |         |          |             |
 2    |  1   |    1   |   1     |     1    |      1      |    1
      |      |        |         |          |             |
 3    |  1   |    4   |   4     |     1    |      1      |    1
      |      |        |         |          |             |
 4    |  1   |    1   |   1     |     3    |      1      |    1
      |      |        |         |          |             |
 5    |  1   |    3   |   3     |     1    |      1      |    1
      |      |        |         |          |             |
 6    |  1   |    1   |   1     |     1    |      1      |    1
      |      |        |         |          |             |
 7    |  1   |    2   |   2     |     1    |      1      |    1
      |      |        |         |          |             |
 8    |  1   |    1   |   1     |     1    |      1      |    1
      |      |        |         |          |             |
 9    |  1   |    1   |   1     |     1    |      5      |    6
      |      |        |         |          |             |
10    |  1   |    1   |   1     |  1 or 5  |      1      |    1
      |      |        |         |          |             |
11    |  1   |    1   |   1     |     1    |      6      |    6
      |      |        |         |          |             |
12    |  1   |    1   |   1     |     1    |      1      | 1 or 6
      |      |        |         |          |             |
13    |  1   |    1   |   1     |     1    |      1      |    1
      |      |        |         |          |             |
      ---------------------------------------------------------------




Lougheed & Rekhter                                             [Page 27]

RFC 1163                          BGP                          June 1990


Appendix 2.  Comparison with RFC 1105

   Minor changes to the RFC1105 Finite State Machine were necessary to
   accommodate the TCP user interface provided by 4.3 BSD.

   The notion of Up/Down/Horizontal relations present in RFC1105 has
   been removed from the protocol.

   The changes in the message format from RFC1105 are as follows:

       1.  The Hold Time field has been removed from the BGP header and
           added to the OPEN message.

       2.  The version field has been removed from the BGP header and
           added to the OPEN message.

       3.  The Link Type field has been removed from the OPEN message.

       4.  The OPEN CONFIRM message has been eliminated and replaced
           with implicit confirmation provided by the KEEPALIVE message.

       5.  The format of the UPDATE message has been changed
           significantly.  New fields were added to the UPDATE message
           to support multiple path attributes.

       6.  The Marker field has been expanded and its role broadened to
           support authentication.

Appendix 3.  TCP options that may be used with BGP

   If a local system TCP user interface supports TCP PUSH function, then
   each BGP message should be transmitted with PUSH flag set.  Setting
   PUSH flag forces BGP messages to be transmitted promptly to the
   receiver.

   If a local system TCP user interface supports setting precedence for
   TCP connection, then the BGP transport connection should be opened
   with precedence set to Internetwork Control (110) value (see also
   [6]).












Lougheed & Rekhter                                             [Page 28]

RFC 1163                          BGP                          June 1990


References

   [1]  Mills, D., "Exterior Gateway Protocol Formal Specification", RFC
        904, BBN, April 1984.

   [2]  Rekhter, Y., "EGP and Policy Based Routing in the New NSFNET
        Backbone", RFC 1092, T.J. Watson Research Center, February 1989.

   [3]  Braun, H-W., "The NSFNET Routing Architecture", RFC 1093,
        MERIT/NSFNET Project, February 1989.

   [4]  Postel, J., "Transmission Control Protocol - DARPA Internet
        Program Protocol Specification", RFC 793, DARPA, September 1981.

   [5]  Honig, J., Katz, D., Mathis, M., Rekhter, Y., and J. Yu,
        "Application of the Border Gateway Protocol in the Internet",
        RFC 1164, Cornell University Theory Center, Merit/NSFNET,
        Pittsburgh Supercomputing Center, IBM, Merit/NSFNET, June 1990.

   [6]  Postel, J., "Internet Protocol - DARPA Internet Program Protocol
        Specification", RFC 791, DARPA, September 1981.

Security Considerations

   Security issues are not discussed in this memo.

Authors' Addresses



   Kirk Lougheed
   cisco Systems, Inc.
   1525 O'Brien Drive
   Menlo Park, CA 94025

   Phone:  (415) 326-1941

   Email:  LOUGHEED@CISCO.COM


   Yakov Rekhter
   T.J. Watson Research Center IBM Corporation
   P.O. Box 218
   Yorktown Heights, NY 10598

   Phone:  (914) 945-3896

   Email:  YAKOV@IBM.COM





Lougheed & Rekhter                                             [Page 29]