Network Working Group                                        D. Cansever
Request for Comments: 2333                        GTE Laboratories, Inc.
Category: Standards Track                                     April 1998


                 NHRP Protocol Applicability Statement

Status of this Memo

   This document specifies an Internet standards track protocol for the
   Internet community, and requests discussion and suggestions for
   improvements.  Please refer to the current edition of the "Internet
   Official Protocol Standards" (STD 1) for the standardization state
   and status of this protocol.  Distribution of this memo is unlimited.

Copyright Notice

   Copyright (C) The Internet Society (1998).  All Rights Reserved.

Abstract

   As required by the Routing Protocol Criteria [RFC 1264], this memo
   discusses the applicability of the Next Hop Resolution Protocol
   (NHRP) in routing of IP datagrams over Non-Broadcast Multiple Access
   (NBMA) networks, such as ATM, SMDS and X.25.

1. Protocol Documents

   The NHRP protocol description is defined in [1].  The NHRP MIB
   description is defined in [2].

2. Introduction

   This document summarizes the key features of NHRP and discusses the
   environments for which the protocol is well suited.  For the purposes
   of description, NHRP can be considered a generalization of Classical
   IP and ARP over ATM which is defined in [3] and of the Transmission
   of IP Datagrams over the SMDS Service, defined in [4].  This
   generalization occurs in 2 distinct directions.

   Firstly, NHRP avoids the need to go through extra hops of routers
   when the Source and Destination belong to different Logical Internet
   Subnets (LIS).  Of course, [3] and [4] specify that when the source
   and destination belong to different LISs, the source station must
   forward data packets to a router that is a member of multiple LISs,
   even though the source and destination stations may be on the same
   logical NBMA network.  If the source and destination stations belong
   to the same logical NBMA network, NHRP provides the source station

   with an inter-LIS address resolution mechanism at the end of which
   both stations can exchange packets without having to use the services
   of intermediate routers.  This feature is also referred to as
   "short-cut" routing.  If the destination station is not part of the
   logical NBMA network, NHRP provides the source with the NBMA address
   of the current egress router towards the destination.

   The second generalization is that NHRP is not specific to a
   particular NBMA technology.  Of course, [3] assumes an ATM network
   and [4] assumes an SMDS network at their respective subnetwork
   layers.

   NHRP is specified for resolving the destination NBMA addresses of IP
   datagrams over IP subnets within a large NBMA cloud.  NHRP has been
   designed to be extensible to network layer protocols other than IP,
   possibly subject to other network layer protocol specific additions.

   As an important application of NHRP, the Multiprotocol Over ATM
   (MPOA) Working Group of the ATM Forum has decided to adopt and to
   integrate NHRP into its MPOA Protocol specification [5].  As such,
   NHRP will be used in resolving the ATM addresses of MPOA packets
   destined outside the originating subnet.

3. Key Features

   NHRP provides a mechanism to obtain the NBMA network address of the
   destination, or of a router along the path to the destination. NHRP
   is not a routing protocol, but may make use of routing information.
   This is further discussed in Section 5.

   The most prominent feature of NHRP is that it avoids extra router
   hops in an NBMA with multiple LISs.  To this goal, NHRP provides the
   source with the NBMA address of the destination, if the destination
   is directly attached to the NBMA. If the destination station is not
   attached to the NBMA, then NHRP provides the source with the NBMA
   address of an exit router that has connectivity to the destination.
   In general, there may be multiple exit routers that have connectivity
   to the destination.  If NHRP uses the services of a dynamic routing
   algorithm in fulfilling its function, which is necessary for robust
   and scalable operation, then the exit router identified by NHRP
   reflects the selection made by the network layer dynamic routing
   protocol.  In general, the selection made by the routing protocol
   would often reflect a desirable attribute, such as identifying the
   exit router that induces the least number of hops in the original
   routed path.

   NHRP is defined for avoiding extra hops in the delivery of IP packets
   with a single destination.  As such, it is not intended for direct
   use in a point-to-multipoint communication setting.  However,
   elements of NHRP may be used in certain multicast scenarios for the
   purpose of providing short cut routing. Such an effort is discussed
   in [6].  In this case, NHRP would avoid intermediate routers in the
   multicast path. The scalability of providing short-cut paths in a
   multicast environment is an open issue.

   NHRP can be used in host-host, host-router and router-host
   communications.  When used in router-router communication, NHRP (as
   defined in [1]) can produce persistent routing loops if the
   underlying routing protocol looses information critical to loop
   suppression. This may occur when there is a change in router metrics
   across the autonomous system boundaries.  NHRP for router-router
   communication that avoids persistent forwarding loops will be
   addressed in a separate document.

   A special case of router-router communication where loops will not
   occur is when the destination host is directly adjacent to the non-
   NBMA interface of the egress router.  If it is believed that the
   adjacency of the destination station to the egress router is a stable
   topological configuration, then NHRP can safely be used in this
   router-router communication scenario.  If the NHRP Request has the Q
   bit set, indicating that the requesting party is a router, and if the
   destination station is directly adjacent to the egress router as a
   stable topological configuration, then the egress router can issue a
   corresponding NHRP reply.  If the destination is not adjacent to the
   egress router, and if Q bit is set in the Request, then a safe mode
   of operation for the egress router would be to issue a negative NHRP
   Reply (NAK) for this particular request, thereby enforce data packets
   to follow the routed path.

   As a result of having inter-LIS address resolution capability, NHRP
   allows the communicating parties to exchange packets by fully
   utilizing the particular features of the NBMA network.  One such
   example is the use of QoS guarantees when the NMBA network is ATM.

   Here, due to short-cut routing, ATM provided QoS guarantees can be
   implemented without having to deal with the issues of re-assembling
   and re-segmenting IP packets at each network layer hop.

   NHRP protocol can be viewed as a client-server interaction.  An NHRP
   Client is the one who issues an NHRP Request. An NHRP Server is the
   one who issues a reply to an NHRP request, or the one who forwards a
   received NHRP request to another Server. Of course, an NHRP entity
   may act both as a Client and a Server.

4. Use of NHRP

   In general, issuing an NHRP request is an application dependent
   action [7].  For applications that do not have particular QoS
   requirements, and that are executed within a short period of time, an
   NBMA short-cut may not be a necessity. In situations where there is a
   "cost" associated with NBMA short-cuts, such applications may be
   better served by network layer hop-by-hop routing. Here, "cost" may
   be understood in a monetary context, or as additional strain on the
   equipment that implements short-cuts. Therefore, there is a trade-off
   between the "cost" of a short-cut path and its utility to the user.
   Reference [7] proposes that this trade-off should be addressed at the
   application level. In an environment consisting of LANs and routers
   that are interconnected via dedicated links, the basic routing
   decision is whether to forward a packet to a router, or to broadcast
   it locally.  Such a decision on local vs. remote is based on the
   destination address. When routing IP packets over an NBMA network,
   where there is potentially a direct Source to Destination
   connectivity with QoS options, the decision on local vs. remote is no
   longer as fundamentally important as in the case where packets have
   to traverse routers that are interconnected via dedicated links.
   Thus, in an NBMA network with QoS options, the basic decision becomes
   the one of short-cut vs. hop-by-hop network layer routing.  In this
   case, the relevant criterion becomes applications' QoS requirements
   [7]. NHRP is particularly applicable for environments where the
   decision on local vs. remote is superseded by the decision on short-
   cut vs. hop-by-hop network layer routing.

   Let us assume that the trade-off is in favor of a short-cut NBMA
   route.  Generally, an NHRP request can be issued by a variety of NHRP
   aware entities, including hosts and routers with NBMA interfaces.  If
   an IP packet traverses multiple hops before a short-cut path has been
   established, then there is a chance that multiple short-cut paths
   could be formed. In order to avoid such an undesirable situation, a
   useful operation rule is to authorize only the following entities to
   issue an NHRP request and to perform short-cut routing.

     i)  The host that originates the IP packet, if the host has an NBMA
         interface.
     ii) The first router along the routing path of the IP packet such
         that the next hop is reachable through the NBMA interface of
         that particular router.
    iii) A policy router within an NBMA network through which the IP
         packet has to traverse.

5. Protocol Scalability

   As previously indicated, NHRP is defined for the delivery of IP
   packets with a single destination. Thus, this discussion is confined
   to a unicast setting.  The scalability of NHRP can be analyzed at
   three distinct levels:

     o Client level
     o LIS level
     o Domain level

   At the the Client level, the scalability of NHRP is affected by the
   processing and memory limitations of the NIC that provides interface
   to the NBMA network.  When the NBMA network is connection oriented,
   such as ATM, NIC limitations may bound the scalability of NHRP in
   certain applications.  For example, a server that handles hundreds of
   requests per second using an ATM interface may be bounded by the
   performance characteristics of the corresponding NIC.  Similarly,
   when the NHRP Client resides at an NBMA interface of a router, memory
   and processing limitations of router's NIC may bound the scalability
   of NHRP.  This is because routers generally deal with an aggregation
   of traffic from multiple sources, which in turn creates a potentially
   large number of SVCCs out of the router's NBMA interface.

   At the LIS level, the main issue is to maintain and deliver a sizable
   number of NBMA to Network layer address mappings within large LISs.
   To this goal, NHRP implementations can use the services of the Server
   Cache Synchronization Protocol (SCSP) [8] that allows multiple
   synchronized NHSs within an LIS, and hence resolve the associated
   scalability issue.

   At the NHRP Domain level, network layer routing is used in resolving
   the NBMA address of a destination outside the LIS.  As such, the
   scalability of NHRP is closely tied to the scalability of the network
   layer routing protocol used by NHRP.  Dynamic network layer routing
   protocols are proven to scale well.  Thus, when used in conjunction
   with dynamic routing algorithms, at the NHRP domain level, NHRP
   should scale in the same order as the routing algorithm, subject to
   the assumption that all the routers along the path are NHRP aware.
   If an NHRP Request is processed by a router that does not implement
   NHRP, it will be silently discarded.  Then, short-cuts cannot be
   implemented and connectivity will be provided on a hop-by-hop basis.

   Thus, when NHRP is implemented in conjunction with dynamic network
   layer routing, a scaling requirement for NHRP is that virtually all
   the routers within a logical NBMA network should be NHRP aware.

   One can also use static routing in conjunction with NHRP.  Then, not
   all the routers in the NBMA network need to be NHRP aware.  That is,
   since the routers that need to process NHRP control messages are
   specified by static routing, routers that are not included in the
   manually defined static paths do not have to be NHRP aware.  Of
   course, static routing does not scale, and if the destination is off
   the NBMA network, then the use of static routing could result in
   persistently suboptimal routes.  Use of static routing also has
   fairly negative failure modes.

6. Discussion

   NHRP does not replace existing routing protocols. In general, routing
   protocols are used to determine the proper path from a source host or
   router, or intermediate router, to a particular destination.  If the
   routing protocol indicates that the proper path is via an interface
   to an NBMA network, then NHRP may be used at the NBMA interface to
   resolve the destination IP address into the corresponding NBMA
   address.  Of course, the use of NHRP is subject to considerations
   discussed in Section 4.

   Assuming that NHRP is applicable and the destination address has been
   resolved, packets are forwarded using the particular data forwarding
   and path determination mechanisms of the underlying NBMA network.
   Here, the sequence of events are such that route determination is
   performed by IP routing, independent of NHRP. Then, NHRP is used to
   create a short-cut track upon the path determined by the IP routing
   protocol. Therefore, NHRP "shortens" the routed path.  NHRP (as
   defined in [1]) is not sufficient to suppress persistent forwarding
   loops when used for router-router communication if the underlying
   routing protocol looses information critical to loop suppression [9].
   Work is in progress [10] to augment NHRP to enable its use for the
   router-router communication without persistent forwarding loops.

   When the routed path keeps changing on some relatively short time
   scale, such as seconds, this situation will have an effect on the
   operation of NHRP. In certain router-router operations, changes in
   the routed path could create persistent routing loops. In host-
   router, or router-host communications, frequent changes in routed
   paths could result in inefficiencies such as frequent creation of
   short-cut paths which are short lived.

7. Security Considerations

   NHRP is an address resolution protocol, and SCSP is a database
   synchronization protocol.  As such, they are possibly subject to
   server (for NHRP) or peer (for SCSP) spoofing and denial of service
   attacks.  They both provide authentication mechanisms to allow their

   use in environments in which spoofing is a concern.  Details can be
   found in sections 5.3.4 in [1] and B.3.1 in [8].  There are no
   additional security constraints or concerns raised in this document
   that are not already discussed in the referenced sections.

References

   [1] Luciani, J., Katz, D., Piscitello, D., Cole, B., and
       N. Doraswamy, "NMBA Next Hop Resolution Protocol (NHRP)", RFC
       2332, April 1998.

   [2] Greene, M., and J. Luciani, "NHRP Management Information Base",
       Work in Progress.


   [3] Laubach, M., and J. Halpern, "Classical IP and ARP over ATM", RFC
       2225, April 1998.

   [4] Lawrance, J., and D. Piscitello, "The Transmission of IP
       datagrams over the SMDS service", RFC 1209, March 1991.

   [5] Multiprotocol Over ATM Version 1.0, ATM Forum Document
       af-mpoa-0087.000

   [6] Rekhter, Y., and D. Farinacci, "Support for Sparse Mode PIM over
       ATM", Work in Progress.

   [7] Rekhter, Y., and D. Kandlur, "Local/Remote" Forwarding Decision
       in Switched Data Link Subnetworks", RFC 1937, May 1996.

   [8] Luciani, J., Armitage, G., Halpern, J., and N. Doraswamy, "Server
       Cache Synchronization Protocol (SCSP) - NBMA", RFC 2334, April
       1998.

   [9] Cole, R., Shur, D., and C. Villamizar, "IP over ATM: A Framework
       Document", RFC 1932, April 1996.

   [10] Rekhter, Y., "NHRP for Destinations off the NBMA Subnetwork",
        Work in Progress.

Acknowledgements

   The author acknowledges valuable contributions and comments from many
   participants of the ION Working Group, in particular from Joel
   Halpern of Newbridge Networks, David Horton of Centre for Information
   Technology Research, Andy Malis of Nexion, Yakov Rekhter and George
   Swallow of Cisco Systems and Curtis Villamizar of ANS.

Author's Address

   Derya H. Cansever
   GTE Laboratories Inc.
   40 Sylvan Rd. MS 51
   Waltham MA 02254

   Phone: +1 617 466 4086
   EMail: dcansever@gte.com

Full Copyright Statement

   Copyright (C) The Internet Society (1998).  All Rights Reserved.

   This document and translations of it may be copied and furnished to
   others, and derivative works that comment on or otherwise explain it
   or assist in its implementation may be prepared, copied, published
   and distributed, in whole or in part, without restriction of any
   kind, provided that the above copyright notice and this paragraph are
   included on all such copies and derivative works.  However, this
   document itself may not be modified in any way, such as by removing
   the copyright notice or references to the Internet Society or other
   Internet organizations, except as needed for the purpose of
   developing Internet standards in which case the procedures for
   copyrights defined in the Internet Standards process must be
   followed, or as required to translate it into languages other than
   English.

   The limited permissions granted above are perpetual and will not be
   revoked by the Internet Society or its successors or assigns.

   This document and the information contained herein is provided on an
   "AS IS" basis and THE INTERNET SOCIETY AND THE INTERNET ENGINEERING
   TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING
   BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION
   HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF
   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.