ROLC Working Group Derya H. Cansever INTERNET DRAFT GTE Laboratories, Inc. February 1996 Expiration Date July 1996 NHRP Protocol Applicability Statement Status of this Memo This document is an Internet Draft. Internet Drafts are working documents of the Internet Engineering Task Force (IETF), its Areas, and its Working Groups. Note that other groups may also distribute working documents as Internet Drafts. Internet Drafts are draft documents valid for a maximum of six months. Internet Drafts may be updated, replaced, or obsoleted by other documents at any time. It is not appropriate to use Internet Drafts as reference material or to cite them other than as a "working draft" or "work in progress." Please check the 1id-abstracts.txt listing contained in the internet-drafts Shadow Directories on nic.ddn.mil, nnsc.nsf.net, nic.nordu.net, ftp.nisc.sri.com, or munnari.oz.au to learn the current status of any Internet Draft. Abstract As required by the Routing Protocol Criteria [RFC 1264], this draft report 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. The final form of this draft report is a prerequisite to advancing NHRP on the standards track. 1. Protocol Documents The NHRP protocol description is defined in [1] in its draft form. The NHRP protocol analysis is documented in TBD [2]. The NHRP MIB description is defined in [3] in its draft form. 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 [4] and of the Transmission of IP Datagrams over the SMDS Service, defined in [5]. 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, [4] and [5] 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 egress router towards the destination. The second generalization is that NHRP is not specific to a particular NBMA technology. Of course, [4] assumes an ATM network and [5] assumes an SMDS network at their respective subnetwork layers. NHRP is specified for resolving the NBMA addresses IP datagrams over large clouds of NBMA networks. In principle, NHRP should also be extensible to network layer protocols other than IP, without major modifications to the specification provided in [1]. 3. Key Features NHRP is not a routing protocol, but an inter-LIS address resolution mechanism that may make use of network layer routing in resolving the NBMA address of the destination. This is further discussed in Section 5. The most prominent feature of NHRP is that it avoids extra 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. 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]. NHRP can be used in host-host, host-router and router-host communications. When used in router-router communication, NHRP can produce persistent routing loops. NHRP for router-router communication will be addressed in 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, possibly with the B bit set, indicating that the information is stable. 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 not to issue an NHRP Reply for this particular request. As a result of 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. NHRP uses network layer routing in resolving the NBMA address of the destination. The related routing tables can be populated using the services of network layer dynamic routing algorithms, or they may be statically configured. If network layer dynamic routing algorithms are used, NHRP Servers would be implemented in a router, or in some device that participates to a network layer dynamic routing algorithm. If static routing is used, then NHRP Servers do not necessarily have to participate to network layer dynamic routing algorithms. All they are required to do is to reply to NHRP Requests using their IP to NBMA address resolution tables, or to forward them to another Server, using some pre-determined forwarding rules. 4. Use of NHRP In general, issuing an NHRP request would be 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 MBMA 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. Then, 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 NHRP uses network layer routing in resolving the NBMA address of the destination. 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, NHRP should scale in the same order as the routing algorithm, subject to two issues. The first issue is related to the memory size and the required processing power for the address resolution tables at the NHSs. The routing algorithm divides the network into moderately sized subnets. Assignment of the areas of responsibility for each NHS in a way similar to the operation of the routing algorithm will resolve the issue of address resolution table size. The second issue is related to the NHRP awareness of the routers. If a router on the routed path of an NHRP Request does not implement NHRP, it will silently discard the Request. 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, 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. Of course, static routing does not scale well. Also, when static routing is used, it may not be possible to forward the IP packet to be transmitted along with the path of the NHRP Request. 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 the 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. An advantage of this approach is that it "shortens" the routed path. A disadvantage is that it may create persistent routing loops when used in router-to-router communication [8]. As noted in Section 3, Router-to-Router NHRP will be addressed in a separate document. References [1] NMBA Next Hop Resolution Protocol (NHRP), Dave Katz, David Piscitello, Bruce Cole and James Luciani, draft-ietf-rolc-nhrp-07.txt. [2] TBD [3] NHRP Management Information Base, M. Patrick, draft-ietf-rolc -nhrp-mib-01.txt [4] Classical IP and ARP over ATM, Mark Laubach, RFC 1577. [5] Transmission of IP datagrams over the SMDS service, J. Lawrance and D. Piscitello, RFC 1209. [6] Support for Sparse Mode PIM over ATM, Yakov Rekhter and Dino Farinacci, draft-rekhter-pim-atm-00.txt [7] "Local/Remote" Forwarding Decision in Switched Data Link Subnetworks, Yakov Rekhter and Dilp Kandlur, draft-ietf-rolc-apr-04.txt. [8] IP over ATM: A Framework Document, R.G. Cole, D.H. Shur and C. Villamizar, draft-ietf-ipatm-framework-doc-07.ps Acknowledgements The author acknowledges valuable contributions and comments from many participants of the ROLC Working Group, in particular from Curtis Villamizar, Yakov Rekhter, Joel Halpern and Andy Malis. Author's Address Derya H. Cansever GTE Laboratories Inc. 40 Sylvan Rd. MS 51 Waltham MA 02254 Phone: +1 617 466 4086 Email: dhc2@gte.com Expiration Date July 1996