Network Working Group Mark Laubach INTERNET DRAFT Com21, Inc. Expires 29 February 1996 31 August 1995 Obsoletes: 1577, 1626 Classical IP and ARP over ATM Update (Part Deux) Status of this Memo This memo 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 lid-abstracts.txt listing contained in the internet-drafts shadow directories on nic.ddn.mil, nnsc.nsf.net, nic.nordu.net, ftp.nisc.src.com, or munnari.oz.au to learn the current status of any Internet Draft. Update Information This draft represents an update to RFC-1577 and RFC-1626. Changes to RFC-1577 are bracketed in the text by "[changebar on]" and "[changebar off]". The following changes/updates are incorporated into this document as per the consensus at the Danvers IETF meeting (see Danvers minutes): o Single ATMARP server address to ATMARP and NHRP server lists o RFC1626 text replaces MTU section o Client registration procedure from In_ATMARP to first ATMARP_Request o Clarification of variable length ATMARP packet format o Clarification of ARP_NAK packet format Laubach [Page 1] DRAFT Classical IP and ARP over ATM Update August 1995 o Clarification of In_ATMARP packet format for null IPv4 addresses o Expanded issues list o Multiple ATMARP server synchronization protocol The following item from the Danvers meeting will incorporated into subsequent I-D's: o Classical RFC1577++ MIB definition Additionally, I've added some requirements for unsupported operation detection and notification in ATMARP servers. Looking at the future, this will be necessary as MARS and other "arp level" services are deployed and made co-resident with ATMARP servers. Once we resolve a few issues, the following may be moved into this document: o RFC1483 LLC/SNAP Encapsulation in the main body o Multicast LLC/SNAP Encapsulation in the main body o The other RFC1483 encapsulations as informational in an appendix. The "Update (Part Deux)" will be removed from the title at a later time, prior to any working group action for this memo. I am trying to locate easily obtainable published references for the Xerox PARC epidemic database work. Some folks there think they can find something to help out. 1. ABSTRACT This memo defines an initial application of classical IP and ARP in an Asynchronous Transfer Mode (ATM) network environment configured as a Logical IP Subnetwork (LIS) as described in Section 5. This memo does not preclude the subsequent development of ATM technology into areas other than a LIS; specifically, as single ATM networks grow to replace many Ethernet local LAN segments and as these networks become globally connected, the application of IP and ARP will be treated differently. This memo considers only the application of ATM as a direct replacement for the "wires" and local LAN segments connecting IP end-stations ("members") and routers operating in the "classical" LAN-based paradigm. Issues raised by MAC level bridging and LAN emu- lation are beyond the scope of this paper. This memo introduces general ATM technology and nomenclature. Laubach [Page 2] DRAFT Classical IP and ARP over ATM Update August 1995 Readers are encouraged to review the ATM Forum and ITU-TS (formerly CCITT) references for more detailed information about ATM implementa- tion agreements and standards. 2. ACKNOWLEDGMENT The author would like to thank the efforts of the IP over ATM Working Group of the IETF. Without their substantial, and sometimes con- tentious support, of the Classical IP over ATM model, this updated memo would not have been possible. Section 7, on Default MTU, has been incorporated directly from Ran Atkinson's RFC-1626, with his permission. Special thanks to Berry Kercheval, Bryan Lyles, and Christoph Schuba from Xerox PARC for their help in refining the server synchronization protocol an infecting me with the epidemic database algorithms. Thanks for Andy Malis for an early review and comments for rolc related issues. 3. CONVENTIONS The following language conventions are used in the items of specifi- cation in this document: o MUST, SHALL, or MANDATORY -- the item is an absolute requirement of the specification. o SHOULD or RECOMMEND -- this item should generally be followed for all but exceptional circumstances. o MAY or OPTIONAL -- the item is truly optional and may be followed or ignored according to the needs of the implementor. 4. INTRODUCTION The goal of this specification is to allow compatible and interopera- ble implementations for transmitting IP datagrams and ATM Address Resolution Protocol (ATMARP) requests and replies over ATM Adaptation Layer 5 (AAL5)[2,6]. [changebar on] This memo specifies the stable foundation baseline operational model which will always be available in IP and ARP over ATM implementa- tions. Subsequent memos will build upon and refine this model, how- ever, in the absence or failure of those extensions, operations will Laubach [Page 3] DRAFT Classical IP and ARP over ATM Update August 1995 default to the specifications contained in this memo. Consequently, this memo will not reference these other extensions. [changebar off] This memo defines only the operation of IP and address resolution over ATM, and is not meant to describe the operation of ATM networks. Any reference to virtual connections, permanent virtual connections, or switched virtual connections applies only to virtual channel con- nections used to support IP and address resolution over ATM, and thus are assumed to be using AAL5. This memo places no restrictions or requirements on virtual connections used for other purposes. Initial deployment of ATM provides a LAN segment replacement for: 1) Local area networks (e.g., Ethernets, Token Rings and FDDI). 2) Local-area backbones between existing (non-ATM) LANs. 3) Dedicated circuits or frame relay PVCs between IP routers. Note: In 1), local IP routers with one or more ATM interfaces will be able to connect islands of ATM networks. In 3), public or private ATM Wide Area networks will be used to connect IP routers, which in turn may or may not connect to local ATM networks. ATM WANs and LANs may be interconnected. [changebar on] Private ATM networks (local or wide area) will use the private ATM address structure specified in the ATM Forum UNI 3.1 specification [9]. This structure is modeled after the format of an OSI Network Service Access Point Address. A private ATM address uniquely identi- fies an ATM endpoint. Public networks will use either the address structure specified in ITU-TS recommendation E.164 or the private network ATM address structure. An E.164 address uniquely identifies an interface to a public network. [changebar off] The characteristics and features of ATM networks are different than those found in LANs: [changebar on] o ATM provides a Virtual Connection (VC) switched environment. VC setup may be done on either a Permanent Virtual Connection (PVC) or dynamic Switched Virtual Connection (SVC) basis. SVC call Laubach [Page 4] DRAFT Classical IP and ARP over ATM Update August 1995 management signalling is performed via implementations of the UNI 3.1 protocol [7,9]. [changebar off] o Data to be passed by a VC is segmented into 53 octet quantities called cells (5 octets of ATM header and 48 octets of data). o The function of mapping user Protocol Data Units (PDUs) into the information field of the ATM cell and vice versa is performed in the ATM Adaptation Layer (AAL). When a VC is created a specific AAL type is associated with the VC. There are four different AAL types, which are referred to individually as "AAL1", "AAL2", "AAL3/4", and "AAL5". (Note: this memo concerns itself with the mapping of IP and ATMARP over AAL5 only. The other AAL types are mentioned for introductory purposes only.) The AAL type is known by the VC end points via the call setup mechanism and is not car- ried in the ATM cell header. For PVCs the AAL type is adminis- tratively configured at the end points when the Connection (cir- cuit) is set up. For SVCs, the AAL type is communicated along the VC path via UNI 3.1 as part of call setup establishment and the end points use the signaled information for configuration. ATM switches generally do not care about the AAL type of VCs. The AAL5 format specifies a packet format with a maximum size of (64K - 1) octets of user data. Cells for an AAL5 PDU are trans- mitted first to last, the last cell indicating the end of the PDU. ATM standards guarantee that on a given VC, cell ordering is preserved end-to-end. NOTE: AAL5 provides a non-assured data transfer service - it is up to higher-level protocols to provide retransmission. o ATM Forum signaling defines point-to-point and point-to- multipoint Connection setup [9]. Multipoint-to-multipoint VCs are not yet specified by ITU-TS or ATM Forum. o An ATM Forum ATM endpoint address is either encoded as an NSAP Address (NSAPA) or is an E.164 Public-UNI address [9]. In some cases, both an ATM endpoint address and an E.164 Public UNI address are needed by an ATMARP client to reach another host or router. Since the use of ATM endpoint addresses and E.164 public UNI addresses by ATMARP are analogous to the use of Ethernet addresses, the notion of "hardware address" is extended to encom- pass ATM addresses in the context of ATMARP, even though ATM addresses need not have hardware significance. ATM Forum NSAPAs use the same basic format as U.S. GOSIP NSAPAs [11]. Note: ATM Forum addresses should not be construed as being U.S. GOSIP NSAPAs. They are not, the administration is different, which fields get filled out are different, etc. Laubach [Page 5] DRAFT Classical IP and ARP over ATM Update August 1995 This memo describes the initial deployment of ATM within "classical" IP networks as a direct replacement for local area networks (Ether- nets) and for IP links which interconnect routers, either within or between administrative domains. The "classical" model here refers to the treatment of the ATM host adapter as a networking interface to the IP protocol stack operating in a LAN-based paradigm. Characteristics of the classical model are: [changebar on] o The same maximum transmission unit (MTU) size is the default for all VCs in a LIS [2]. However, on a VC-by-VC point-to-point basis, the MTU size may be negotiated during connection setup using Path MTU Discovery to better suit the needs of the cooper- ating pair of IP members or the attributes of the communications path. (Refer to Section 7.3) [changebar off] o Default LLC/SNAP encapsulation of IP packets. o End-to-end IP routing architecture stays the same. o IP addresses are resolved to ATM addresses by use of an ATMARP service within the LIS - ATMARPs stay within the LIS. From a client's perspective, the ATMARP architecture stays faithful to the basic ARP model presented in [3]. o One IP subnet is used for many hosts and routers. Each VC directly connects two IP members within the same LIS. Future memos will describe the operation of IP over ATM when ATM net- works become globally deployed and interconnected. The deployment of ATM into the Internet community is just beginning and will take many years to complete. During the early part of this period, we expect deployment to follow traditional IP subnet bound- aries for the following reasons: o Administrators and managers of IP subnetworks will tend to ini- tially follow the same models as they currently have deployed. The mindset of the community will change slowly over time as ATM increases its coverage and builds its credibility. o Policy administration practices rely on the security, access, routing, and filtering capability of IP Internet gateways: i.e. firewalls. ATM will not be allowed to "back-door" around these mechanisms until ATM provides better management capability than Laubach [Page 6] DRAFT Classical IP and ARP over ATM Update August 1995 the existing services and practices. o Standards for global IP over ATM will take some time to complete and deploy. This memo details the treatment of the classical model of IP and ATMARP over ATM. This memo does not preclude the subsequent treatment of ATM networks within the IP framework as ATM becomes globally deployed and interconnected; this will be the subject of future docu- ments. This memo does not address issues related to transparent data link layer interoperability. 5. IP SUBNETWORK CONFIGURATION [changebar on] 5.1 Background In the LIS scenario, each separate administrative entity configures its hosts and routers within a LIS. Each LIS operates and communi- cates independently of other LISs on the same ATM network. In the classical model, hosts communicate directly via ATM to other hosts within the same LIS using the ATMARP service as the mechanism for resolving target IP addresses to target ATM endpoint addresses. The ATMARP service has LIS scope only and serves all hosts in the LIS. Communication to hosts located outside of the local LIS is pro- vided via an IP router. This router is an ATM endpoint attached to the ATM network that is configured as a member of one or more LISs. This configuration MAY result in a number of disjoint LISs operating over the same ATM network. Using this model hosts of differing IP subnets MUST communicate via an intermediate IP router even though it may be possible to open a direct VC between the two IP members over the ATM network. Using a classical plus next hop address resolution model, the ATMARP service is supplemented with a Next Hop Resolution Protocol (NHRP) service [18]. The NHRP service provides an IP to ATM endpoint address resolution service for target hosts located outside the LIS. With NHRP a classical host, after not receiving a satisfactory reply from the ATMARP service, may then query the NHRP service for resolution. If the target IP address is known to the NHRP service an ATM endpoint address is returned for the query. The host may then open an ATM con- nection to that host. This configuration allows IP members to "cut through" the classical LIS boundaries on a case by case basis in favor of more direct ATM connections between cooperating IP members. Laubach [Page 7] DRAFT Classical IP and ARP over ATM Update August 1995 By default, the ATMARP service and the classical LIS routing model MUST be available to any IP member in the LIS. 5.2 LIS Configuration Requirements [changebar off] The requirements for IP members (hosts, routers) operating in an ATM LIS configuration are: o All members of the LIS have the same IP network/subnet number and address mask [8]. o All members within a LIS are directly connected to the ATM net- work. o All members of a LIS MUST have a mechanism for resolving IP addresses to ATM addresses via ATMARP (based on [3]) and vice versa via InATMARP (based on [12]) when using SVCs. Refer to Section 8 "Address Resolution" in this memo. o All members of a LIS MUST have a mechanism for resolving VCs to IP addresses via InATMARP (based on [12]) when using PVCs. Refer to Section 8 "Address Resolution" in this memo. o All members within a LIS MUST be able to communicate via ATM with all other members in the same LIS; i.e., the Virtual Connection topology underlying the intercommunication among the members is fully meshed. [changebar on] o Members of a LIS MAY but are NOT REQUIRED to have a mechanism for resolving IP addresses to ATM addresses via NHRP (based on [18]). o When NHRP is not available to a LIS member, all IP stations located outside of the LIS are accessed via a router. [changebar off] The following list identifies the set of ATM specific parameters that MUST be implemented in each IP station connected to the ATM network: o ATM Hardware Address (atm$ha). The ATM address of the individual IP station. [changebar on] Laubach [Page 8] DRAFT Classical IP and ARP over ATM Update August 1995 o ATMARP Request Address list (atm$arp-req-list): atm$arp-req-list is a list containing one or more ATM addresses of individual ATMARP servers located within the LIS. In an SVC environment, ATMARP servers are used to resolve target IP addresses to target ATM address via an ATMARP request and reply protocol. ATMARP servers MUST have authoritative responsibility for resolving ATMARP requests of all IP members using SVCs located within the LIS. Implementations MUST support a minimum list size of at least three entries which contain either null or non-null addresses. o NHS Request Address list (atm$nhs-req-list): atm$nhs-req-list is a list containing one or more ATM addresses of individual Next Hop Resolution Protocol (NHRP) servers also called Next Hop Servers (NHS) [18]. In an SVC environment, NHRP servers are used to resolve target IP addresses to target ATM address via an NHRP request and reply protocol. The NHS is typically used to resolve IP addresses to ATM address for IP addresses located outside the LIS. Implementations MUST support a minimum list size of at least three entries which contain either null or non-null entries. Implementations MUST be able to identify (type) the list entry as a unicast or multicast ATM address. Note: the minimum default lists sizes for atm$arp-req-list and atm$nhs-req-list were selected to enforce implementation interoper- ability and testing for these minimum sizes and to allow network designers to have a priori knowledge of the minimum capabilities of implementations conforming to this specification. A LIS MUST have a single ATMARP service (single server or multiple synchronized servers) configured and available to all members of the LIS who use SVCs. If the ATMARP service for the LIS is being provided by a synchronized multiple-server ATMARP service implementation (see Section 8.10) then each client MUST be configured with at least one non-null entry in atm$arp-req-list specifying one of the synchronized servers otherwise there are no requirements on the number or order of the entries in atm$arp-req-list and there are no requirements that all clients be configured identically. If the ATMARP service is being provided by a single ATMARP server, then all clients within the LIS MUST be configured identically to have only one non-null entry in atm$arp-req-list configured with the same address of the single ATMARP server. If the IP member is operating with PVCs only, then the atm$arp-req- list and the atm$nhs-req-list MUST be configured with all null entries and the client MUST not make queries to either address Laubach [Page 9] DRAFT Classical IP and ARP over ATM Update August 1995 resolution service. The classical IP member MAY allow the optional use of the supplemen- tal NHRP address resolution service on a client-by-client basis if the local administration has provided the service; i.e., it is not necessary for all IP members in the LIS to be configured to use NHRP as the default IP router path with be always available. The guide- lines for coordinating ATMARP and NHRP service queries are detailed in Section 8.5. It is beyond the scope of this memo to detail how the NHS is configured and administrated. Within the restrictions mentioned above and in Section 8, local administration MUST decide which server address(es) are appropriate for atm$arp-req-list and atm$nhs-req-list. Manual or automatic configuration of the addresses and address lists presented in this section is implementation dependent and beyond the scope of this document; i.e. this memo does not require any specific configuration method. This memo does require that these addresses MUST be configured completely on the client, as appropriate for the LIS, prior to use by any service or operation detailed in this memo. 5.3 LIS Router Additional Configuration [changebar off] It is RECOMMENDED that routers providing LIS functionality over the ATM network also support the ability to interconnect multiple LISs. Routers that wish to provide interconnection of differing LISs MUST be able to support multiple sets of these parameters (one set for each connected LIS) and be able to associate each set of parameters to a specific IP network/ subnet number. In addition, it is RECOM- MENDED that a router be able to provide this multiple LIS support with a single physical ATM interface that may have one or more indi- vidual ATM endpoint addresses. Note: this does not necessarily mean different End System Identifiers (ESIs) when NSAPAs are used. The last octet of an NSAPA is the NSAPA Selector (SEL) field which can be used to differentiate up to 256 different LISs for the same ESI. (Refer to Section 5.1.3.1, "Private Networks" in [9].) 6. IP PACKET FORMAT Implementations MUST support IEEE 802.2 LLC/SNAP encapsulation as described in [2]. LLC/SNAP encapsulation is the default packet for- mat for IP datagrams. This memo recognizes that other encapsulation methods may be used Laubach [Page 10] DRAFT Classical IP and ARP over ATM Update August 1995 however, in the absence of other knowledge or agreement, LLC/SNAP encapsulation is the default. [changebar on] This memo recognizes that end-to-end signaling within ATM will allow negotiation of encapsulation method on a per-VC basis. [changebar off] 7. Default Value for IP MTU over ATM AAL5 Protocols in wide use throughout the Internet, such as the Network File System (NFS), currently use large frame sizes (e.g. 8 KB). Empirical evidence with various applications over the Transmission Control Protocol (TCP) indicates that larger Maximum Transmission Unit (MTU) sizes for the Internet Protocol (IP) tend to give better performance. Fragmentation of IP datagrams is known to be highly undesirable [16]. It is desirable to reduce fragmentation in the net- work and thereby enhance performance by having the IP Maximum Trans- mission Unit (MTU) for AAL5 be reasonably large. NFS defaults to an 8192 byte frame size. Allowing for RPC/XDR, UDP, IP, and LLC headers, NFS would prefer a default MTU of at least 8300 octets. Routers can sometimes perform better with larger packet sizes because most of the performance costs in routers relate to "packets handled" rather than "bytes transferred". So there are a number of good reasons to have a reasonably large default MTU value for IP over ATM AAL5. RFC-1209 specifies the IP MTU over SMDS to be 9180 octets, which is larger than 8300 octets but still in the same range [1]. There is no good reason for the default MTU of IP over ATM AAL5 to be different from IP over SMDS, given that they will be the same magnitude. Having the two be the same size will be helpful in interoperability and will also help reduce incidence of IP fragmentation. Therefore, the default IP MTU for use with ATM AAL5 shall be 9180 octets. All implementations compliant and conformant with this spec- ification shall support at least the default IP MTU value for use over ATM AAL5. 7.1 Permanent Virtual Circuits Implementations which only support Permanent Virtual Circuits (PVCs) will (by definition) not implement any ATM signalling protocol. Such implementations shall use the default IP MTU value of 9180 octets unless both parties have agreed in advance to use some other IP MTU value via some mechanism not specified here. Laubach [Page 11] DRAFT Classical IP and ARP over ATM Update August 1995 7.2 Switched Virtual Circuits Implementations that support Switched Virtual Circuits (SVCs) MUST attempt to negotiate the AAL CPCS-SDU size using the ATM signalling protocol. The industry standard ATM signalling protocol uses two dif- ferent parts of the Information Element named "AAL Parameters" to exchange information on the MTU over the ATM circuit being setup [9]. The Forward Maximum CPCS-SDU Size field contains the value over the path from the calling party to the called party. The Backwards Maxi- mum CPCS-SDU Size Identifier field contains the value over the path from the called party to the calling party. The ATM Forum specifies the valid values of this identifier as 1 to 65535 inclusive. Note that the ATM Forum's User-to-Network-Interface (UNI) signalling per- mits the MTU in one direction to be different from the MTU in the opposite direction, so the Forward Maximum CPCS-SDU Size Identifier might have a different value from the Backwards Maximum CPCS-SDU Size Identifier on the same connection. If the calling party wishes to use the default MTU it shall still include the "AAL Parameters" information element with the default values for the Maximum CPCS-SDU Size as part of the SETUP message of the ATM signalling protocol [9]. If the calling party desires to use a different value than the default, it shall include the "AAL Parame- ters" information element with the desired value for the Maximum CPCS-SDU Size as part of the SETUP message of the ATM Signalling Pro- tocol. The called party will respond using the same information ele- ments and identifiers in its CONNECT message response [9]. If the called party receives a SETUP message containing the "Maximum CPCS-SDU Size" in the AAL Parameters information element, it shall handle the Forward and Backward Maximum CPCS-SDU Size Identifier as follows: a) If it is able to accept the ATM MTU values proposed by the SETUP message, it shall include an AAL Parameters information element in its response. The Forward and Backwards Maximum CPCS-SDU Size fields shall be present and their values shall be equal to the corresponding values in the SETUP message. b) If it wishes a smaller ATM MTU size than that proposed, then it shall set the values of the Maximum CPCS-SDU Size in the AAL Parameters information elements equal to the desired value in the CONNECT message responding to the original SETUP message. c) If the calling endpoint receives a CONNECT message that does not contain the AAL Parameters Information Element, but the corre- sponding SETUP message did contain the AAL Parameters Information element (including the forward and backward CPCS-SDU Size Laubach [Page 12] DRAFT Classical IP and ARP over ATM Update August 1995 fields), it shall clear the call with cause "AAL Parameters can- not be supported". d) If either endpoint receives a STATUS message with cause "Informa- tion Element Non-existent or Not Implemented" or cause ""Access Information Discarded", and with a diagnostic field indicating the AAL Parameters Information Element identifier, it shall clear the call with cause "AAL Parameters cannot be supported." e) If either endpoint receives CPCS-SDUs in excess of the negotiated MTU size, it may use IP fragmentation or may clear the call with cause "AAL Parameters cannot be supported". In this case, an error has occurred either due to a fault in an end system or in the ATM network. The error should be noted by ATM network man- agement for human examination and intervention. If the called endpoint incorrectly includes the Forward and Backward Maximum CPCS-SDU Size fields in the CONNECT messages (e.g. because the original SETUP message did not include these fields) or it sets these fields to an invalid value, then the calling party shall clear the call with cause "Invalid Information Element Contents". 7.3 Path MTU Discovery Required The Path MTU Discovery mechanism is Internet Standard RFC-1191 [17] and is an important mechanism for reducing IP fragmentation in the Internet. This mechanism is particularly important because new subnet ATM uses a default MTU sizes significantly different from older sub- net technologies such as Ethernet and FDDI. In order to ensure good performance throughout the Internet and also to permit IP to take full advantage of the potentially larger IP datagram sizes supported by ATM, all routers implementations that comply or conform with this specification must also implement the IP Path MTU Discovery mechanism as defined in RFC-1191 and clarified by RFC-1435 [15]. Host implementations should implement the IP Path MTU Discovery mechanism as defined in RFC-1191. 8. LIS ADDRESS RESOLUTION SERVICES 8.1 ATM-based ARP and InARP Equivalent Services Address resolution within an ATM LIS SHALL make use of the ATM Address Resolution Protocol (ATMARP) (based on [3]) and the Inverse ATM Address Resolution Protocol (InATMARP) (based on [12]) and as defined in this memo. ATMARP is the same protocol as the ARP proto- col presented in [3] with extensions needed to support address reso- lution in a unicast server ATM environment. InATMARP is the same Laubach [Page 13] DRAFT Classical IP and ARP over ATM Update August 1995 protocol as the original InARP protocol presented in [12] but applied to ATM networks. All IP stations MUST support these protocols as updated and extended in this memo. Use of these protocols differs depending on whether PVCs or SVCs are used. 8.2 Permanent Virtual Connections An IP station MUST have a mechanism (e.g. manual configuration) for determining what PVCs it has, and in particular which PVCs are being used with LLC/SNAP encapsulation. The details of the mechanism are beyond the scope of this memo. [changebar on] All IP members supporting PVCs are required to use the Inverse ATM Address Resolution Protocol (InATMARP) (refer to [12]) on those VCs using LLC/SNAP encapsulation. In a strict PVC environment, the receiver SHALL infer the relevant VC from the VC on which the InAT- MARP_Request or response InATMARP_REPLY was received. When the ATM source and/or target address is unknown, the corresponding ATM address length in the InATMARP packet MUST be set to zero (0) indi- cating a null length, and no storage be allocated in the InATMARP packet, otherwise the appropriate address field should be filled in and the corresponding length set appropriately. InATMARP packet for- mat details are presented later in this memo. Directly from [12]: "When the requesting station receives the In[ATM]ARP_Reply, it may complete the [ATM]ARP table entry and use the provided address information. Note: as with [ATM]ARP, information learned via In[ATM]ARP may be aged or invalidated under certain cir- cumstances." IP stations supporting PVCs MUST re-validate ATMARP table entries as part of the table aging process. See the Section 8.5.1 "Client ATMARP Table Aging". [changebar off] 8.3 Switched Virtual Connections [changebar on] SVCs require support from address resolution services for resolving target IP addresses to target ATM endpoint addresses. All members in the LIS MUST use the same service. This service MUST have authorita- tive responsibility for resolving the ATMARP requests of all IP mem- bers within the LIS. ATMARP servers do not actively establish connections. They depend on the clients in the LIS to initiate connections for the ATMARP Laubach [Page 14] DRAFT Classical IP and ARP over ATM Update August 1995 registration procedure and for transmitting ATMARP requests. An indi- vidual client connects to the ATMARP server using a point-to-point LLC/SNAP VC. The client sends normal ATMARP request packets to the server. The ATMARP server examines each ATMARP_Request packet for the source protocol and source hardware address information of the send- ing client and uses this information to build its ATMARP table cache. This information is used to generate replies to any ATMARP requests it receives. InATMARP_Request packets MUST specify valid address information for ATM source number, ATM target number, and source protocol address; i.e., these fields MUST be non-null in InATMARP_Request packets. This memo defines client-server interaction by using a single server approach as a reference model. The single server approach is then expanded into a multiple-server model by the specification of a server-to-server peer-to-peer synchronization protocol as defined in Section 8.9. [changebar off] 8.4 ATMARP Single Server Operational Requirements [changebar on] A single ATMARP server accepts ATM calls/connections from other ATM end points. After receiving any ATMARP_Request, the server will examine the source and target address information in the packet and make note of the VC on which the ATMARP_Request arrived. It will use this information as necessary to build and update its ATMARP table entries. For each ATMARP_Request, then: 1. If the source IP protocol address is the same as the target IP protocol address and a table entry exists for that IP address and if the source ATM hardware address does not match the table entry ATM address and there is an open VC associated with that table entry that is not the same as the VC associated with the ATMARP_Request, the server MUST return the table entry informa- tion in the ATMARP_Reply, and MUST raise a "duplicate IP address detected" condition to the server's management. The table entry is not updated. 2. Otherwise, if the source IP protocol address is the same as the target IP protocol address, and either there is no table entry for that IP address, or a table entry exists for that IP address Laubach [Page 15] DRAFT Classical IP and ARP over ATM Update August 1995 and there is no open VC associated with that table entry, or if the VC associated with that entry is the same as the VC for the ATMARP_Request, the server MUST either create a new entry or update the old entry as appropriate and returns that table entry information in the ATMARP Reply. 3. Otherwise, when the source IP protocol address does not match the target IP protocol address, the ATMARP server will generate the corresponding ATMARP_Reply if it has an entry for the target information in its ATMARP table. Otherwise it will generate a negative ATMARP reply (ATMARP_NAK). 4. Additionally, when the source IP protocol address does not match the target IP protocol address and when the server receives an ATMARP_Request over a VC, where the source IP and ATM address do not have a corresponding table entry, the ATMARP server MUST cre- ate a new table entry for the source information. Explanation: this allows old RFC1577 clients to register with this ATMARP ser- vice by just issuing requests to it. 5. Additionally, when the source IP protocol address does not match the target IP protocol address and where the source IP and ATM addresses match the association already in the ATMARP table and the ATM address matches that associated with the VC, the server MUST update the table timeout on the source ATMARP table entry but only if it has been more than 10 minutes since the last update. Explanation: if the client is sending ATMARP requests to the server over the same VC that it used to register its ATMARP entry, the server should examine the ATMARP request and note that the client is still "alive" by updating the timeout on the client's ATMARP table entry. However, the server should not update the entry for every ATMARP_Request that has been received as this effects the rumor mongering churn in the server synchro- nization system presented later in this section. An ATMARP server MUST have knowledge of any open VCs it has and their association with an ATMARP table entry, and in particular, which VCs support LLC/SNAP encapsulation. In normal operation, active ATMARP clients will revalidate their entries prior to the server aging pro- cess taking effect. Server ATMARP table entries are valid for 20 minutes. If an entry ages beyond 20 minutes without being updated, that entry is deleted from the table regardless of the state of any VCs that may be associ- ated with that entry. [changebar off] Laubach [Page 16] DRAFT Classical IP and ARP over ATM Update August 1995 8.5 ATMARP Client Operational Requirements [changebar on] The ATMARP client is responsible for contacting the ATMARP service to register its own ATMARP information. The client is also responsible for using the ATMARP service and optionally the NHRP service to gain and refresh entry/information about other IP members (server selec- tion overview is discussed in Section 8.6). As noted in Section 5.2, ATMARP clients MUST be configured with the ATM address(es) of the appropriate servers prior to operation. IP clients MUST register their ATM endpoint address with their ATMARP server using the ATM address structure appropriate for their ATM net- work connection: i.e., LISs implemented over ATM LANs following ATM Forum UNI 3.1 should register using Structure 1; LISs implemented over an E.164 "public" ATM network should register using Structure 2. A LIS implemented over a combination of ATM LANs and public ATM net- works may need to register using Structure 3. Implementations based on this memo MUST support all three ATM address structures. See Sec- tion 8.7.1 for more details regarding the ATMARP Request packet for- mat. For operation, clients MUST: 1. Initiate the LLC/SNAP VC connection to a server in the ATMARP service for transmitting and receiving ATMARP packets (server selection is discussed in the Section 8.5). 2. Immediately after opening a successful connection to the ATMARP service, the client MUST transmit an ATMARP_Request packet, requesting a target ATM address for its own IP address as the target IP protocol address. The client checks the ATMARP_Reply and if the source hardware and protocol addresses match the respective target hardware and protocol addresses, the client is registered with the ATMARP service. If the addresses do not match, the client MAY take action, raise alarms, etc.; however, these actions are beyond the scope of this memo. 3. Clients MUST respond to ATMARP_Request and InATMARP_Request pack- ets received on any VC appropriately. (Refer to Section 7, "Pro- tocol Operation" in RFC1293 [12].) 4. Generate and transmit address resolution request packets to the address resolution server of choice (the selection of the server/service is detailed in Section 8.5). Respond to address resolution reply packets appropriately to build/refresh its own client ATMARP table entries. Laubach [Page 17] DRAFT Classical IP and ARP over ATM Update August 1995 5. Generate and transmit InATMARP_Request packets as needed and to process InATMARP_Reply packets appropriately. InATMARP_Reply packets should be used to build/refresh its own client ATMARP table entries. (Refer to Section 7, "Protocol Operation" in [12].) If the client does not maintain an open VC to the server, the client MUST refresh its ATMARP information with the server at least once every 15 minutes. This is done by repeating steps 1 and 2. An ATMARP client MUST have knowledge of any open VCs it has (perma- nent or switched), their association with an ATMARP table entry, and in particular, which VCs support LLC/SNAP encapsulation. 8.5.1 Client ATMARP Table Aging Client ATMARP table entries are valid for a maximum time of 15 min- utes. When an ATMARP table entry ages, an ATMARP client MUST invalidate the table entry. If there is no open VC associated with the invalidated entry, that entry is deleted. In the case of an invalidated entry and an open VC, the client MUST revalidate the entry prior to transmit- ting any non address resolution traffic on that VC; this requirement applies to both PVCs and SVCs. In the case of an open PVC, the client revalidates the entry by transmitting an InATMARP_REQUEST and updating the entry on receipt of an InATMARP_REPLY. In the case of an open SVC, the client revalidates the entry by querying the address resolution service. If a valid reply is received (e.g., ATMARP_Reply), the entry is updated. If the address resolution service cannot resolve the entry (i.e., "host not found"), the SVC should be closed and the associated table entry removed. If the address resolution service is not available (i.e. "server fail- ure") and if the SVC is LLC/SNAP encapsulated, the client MUST attempt to revalidate the entry by transmitting an InATMARP_Request on that VC and updating the entry on receipt of an InATMARP_Reply. If the InATMARP_Request attempt fails to return an InATMARP_Reply, the SVC should be closed and the associated table entry removed. If a VC with an associated invalidated ATMARP table entry is closed, that table entry is removed. Laubach [Page 18] DRAFT Classical IP and ARP over ATM Update August 1995 8.5.2 Non-Normal VC Operations The specific details on client procedures for detecting non-normal VC connection establishment or closures, or failed communications on an established VC are beyond the scope of this memo. It is REQUIRED how- ever, that the client MUST remove the associated ATMARP entry for a VC that fails to operate properly, as defined by the client, when the client closes that VC, when it releases its resources for a VC, or prior to any attempt to reopen that VC. This behavior specifically REQUIRES that the client MUST refresh its ATMARP table information prior to any attempt to re-establish communication to an IP member after a non-normal communications problem has occurred on a VC previ- ously to that IP member. 8.6 Address Resolution Server Selection A client has two address resolution server lists: the ATMARP server list and the NHRP server list. If the client supports PVCs only, the lists are both empty and the client MUST not generate any address resolution requests other than the InATMARP requests on a PVC needed to validate that PVC. If the client supports SVCs, then the client MUST have at least one entry in the ATMARP server list atm$arp-req- list pointing to a server in the ATMARP service. The specific requirements are presented in Section 5.2. For ATMARP address resolution, the client MUST only select/register with a server listed in atm$arp-req-list. For each address resolution request, the following general procedural guidelines MUST be followed: 1. The client MUST start with the ATMARP server list. 2. The client is looking for an ATMARP_Reply to satisfy the ATMARP_Request An ATMARP_Reply terminates the address resolution request process. A receipt of an ATMARP_NAK or an abnormal condi- tion (e.g., server request timeout or non-normal close) causes the client to try another available server in the same list. The client MUST register itself with each new server as per the reg- istration instructions in Section 8.5 under Item 2 prior to issu- ing the ATMARP_Request it is trying to resolve. The client MUST query all available servers in the ATMARP list before moving the NHRP server list. 3. The client is looking for an NHRP Reply to satisfy the NHRP_Request. A NRHP_Reply terminates the address resolution request process. A receipt of a NHRP_NAK or an abnormal condition Laubach [Page 19] DRAFT Classical IP and ARP over ATM Update August 1995 (e.g., server request timeout or non-normal close) causes the client to try another available server in the same list. When all available servers in the NHRP list have been tried, the client terminates server querying. 4. If at least one negative response was received during the lookup process (i.e., ATMARP_NAK or NHRP_NAK) the address resolution process MUST terminate with a "host not found" condition, other- wise it MUST terminate with a "server failure" condition. The client MUST only have one address resolution service VC open at any given time. The details of selecting a server within a list, the marking a server as available or unavailable, and the processing of "host not found" or "server failure" conditions is implementation dependent and beyond the scope of this memo. 8.7 ATMARP Server Synchronization 8.7.1 Goals This section details the extensions to the single ATMARP server model to support a server-to-server neighbor synchronization protocol and operations necessary for a reliable multiple-server ATMARP address resolution service. The multiple-server protocol is based on the fol- lowing design goals: o The service MUST support arbitrary server connection topologies. o Old RFC1577 clients MUST be able to use the multi-server service with the understanding that they are limited to connect to only one server in the service. o Clients, as defined in this memo, MUST be able to use any server in the address resolution service for address registration and ATMARP_Request query resolution. o The model MUST reliably support Class C size IP subnetworks and allow future extensions to support arbitrarily large IP subnet- works. o ATMARP_Requests MUST be allowed to be arbitrarily distributed to any server in the service and not sent directly or indirectly to a single server. This allows the size of the LIS to scale inde- pendently from single server performance. Laubach [Page 20] DRAFT Classical IP and ARP over ATM Update August 1995 o The model MUST minimize the impact of server-to-server synchro- nization messages to a reasonable percentage of the total network load. o The design of the synchronization protocol MUST be based on existing techniques. To meet the above goals, the server synchronization protocol is based on the Xerox PARC epidemic database replication algorithms as described in [14]. In particular, the protocol and operational model makes use of both the "anti-entropy" and "rumor mongering" replica- tion techniques. Rumor mongering is based on the Counter and Feedback mechanism. The general mechanism is as follows: an arbitrary number of servers are configured to provide the ATMARP service for a LIS. Servers are configured to know about one or more other servers and they maintain open VCs with these servers. Clients are configured to know about one or more servers. When a client registers or updates with a server, that server becomes "infective", declares the information as "hot", and randomly spreads the rumor to other "susceptible" neighbor servers that it knows about using an epidemic rumor mongering tech- nique. The infective server keeps a counter of the number of times a susceptible neighbor has already heard the rumor. If the infective server has exhausted its tries to its neighbors or more than a threshold number of neighbors have already heard the rumor, it declares the rumor has "not hot" and ceases to infect its neighbors. When a server receives a rumor that either creates or updates an entry in its ATMARP table, it will mark that entry as "hot", become infective, and begin rumor mongering to its other neighbors. VC state between cooperating servers is used to determine when to initiate the anti-entropy epidemic method. When a server opens a VC to a neighbor server or when a server receives a VC connection from a neighbor, it enters the anti-entropy state. The server then trans- mits to the neighbor a rumor for every entry in its ATMARP table. Any information received from the neighbor server or other server during this process which creates or updates an entry, is marked as hot and queued. All hot rumors are processed only after the anti- entropy process is complete, that is, a server rumors each entry its ATMARP table, concludes, and then goes into normal rumor mongering for any hot rumors it has received after the anti-entropy process was started. Rumors carry unique origination identification based on the originat- ing server's ATM address and an origination timestamp. Servers will only accept rumors from servers it has been configured to neighbor with. Each rumor message is serialized and receiving servers must Laubach [Page 21] DRAFT Classical IP and ARP over ATM Update August 1995 respond with an acknowledgment and a flag stating whether the rumor has already been heard or not. It is assumed that an ATMARP client MUST make requests to only one ATMARP server at any time. Client mobility is detected based on the originating server's ID information which accompanies a rumor. Epidemic techniques are statistical in nature and produce near com- plete rumor propagation coverage when the value of the threshold counter is greater then 1 (one). For value greater than 1, propor- tional redundant traffic is induced into the network. Xerox PARC studies indicate that a value of 2 or 3 is sufficient to induce near complete coverage [14]. 8.7.2 Baseline Operational Requirements Any ATMARP server implemented to the specifications detailed in this memo MUST support synchronized server operation regardless of how deployed in a LIS; i.e. a single ATMARP server for a LIS MUST still be capable of supporting synchronized server operation even if that capability goes unused. For the purposes of this memo, an ATMARP server "neighbor" is another ATMARP server which has been configured to participate in the ATMARP server synchronization operation. It it beyond the scope of this memo to specify how an ATMARP server is configured with the list of its neighbors. In addition to the ATMARP server operation rules presented in Section 8.3, the additional basic operation rules for synchronized server operation are as follows: 1. An ATMARP server MUST maintain a list of the ATM addresses of its neighbor ATMARP servers. Implementations MUST support a minimum list size of at least five entries which contain either null or non-null addresses. 2. An ATMARP server MUST have knowledge of any LLC/SNAP VCs it has to it's known ATMARP server neighbors, and the state of those VCs. 3. An ATMARP server MUST recognize on which VCs it receives an ATMARP_Request or ATMARP_Sync message. 4. An ATMARP server MUST discard any ATMARP_Sync messages that are received on an LLC/SNAP VC which is not associated with one of its neighbors. 5. If an ATMARP server transmits an ATMARP_Sync rumor message on a Laubach [Page 22] DRAFT Classical IP and ARP over ATM Update August 1995 VC to a neighbor, then that VC MUST only be used for ATMARP server-to-server messages. Likewise, if an ATMARP server receives an ATMARP_Sync message on a VC from a neighbor, it MUST assume that that VC is to be used only for ATMARP server-to- server neighbor messages. 6. An ATMARP server MUST be able to detect if it has two or more ATMARP_Sync message VCs (as described in the preceding guideline) open between itself and one of its neighbors. 7. Each ATMARP table entry MUST also record the "origin address" ATM Address and the "origin timestamp". 8. If the receipt of an ATMARP_Request (as detailed in Section 8.3) causes a table entry to be created, that entry is marked as "hot", the origin address is set to the ATMARP server's ATM address, the origin timestamp and the entry local timestamp are each initialized. 9. If the receipt of an ATMARP_Request (as detailed in Section 8.3) causes a table entry to be updated (and the time since the last update or creation is greater than 10 minutes), that entry is marked as "hot", the origin address is set to the ATMARP server's ATM address, the origin timestamp and the entry local timestamp are each updated (refreshed). 10. An ATMARP server MUST have a mechanism for generating unique ATMARP_Sync message serial numbers. 11. An ATMARP server MUST have a mechanism for generating a 32-bit unsigned integer "timestamp" based upon a local mechanism. The resolution of the timestamp must be sufficient to uniquely mark ATMARP table entries that are created on that server via an ATMARP_Request. This timestamp function is referred to as the "origin timestamp" later in this section. 12. If all previously open ATMARP_Sync messages VCs between an ATMARP server and a neighbor have all been closed, for any reason, that neighbor is marked as "down". The details on restarting connec- tions to neighbors is presented later in this section. 8.7.3 Server Startup Operation The following guidelines define server startup operation: 1. When an ATMARP server begins initial operation, it examines its ATMARP neighbor list and opens an LLC/SNAP VC to each listed neighbor. It is implementation dependent and beyond the scope of Laubach [Page 23] DRAFT Classical IP and ARP over ATM Update August 1995 this memo to detail whether this is a serial or parallel opera- tion for the ATMARP server. 2. After successfully opening a VC to a neighbor, the server initi- ates an "anti-entropy" update. That is, for each "non hot" ATMARP table entry, the server transmits a uniquely serialized ATMARP_Sync ATMARP_Rumor message containing that table entry to the neighbor and then waits for an ATMARP_Sync acknowledgment for that serial number. If an ATMARP_Sync ACK_NotHeard or ACK_Heard is received, any feedback is ignored and the sending server moves to the next entry. If an ATMARP_Sync acknowledgement is not received within 5 seconds, the sending server marks the neighbor as "down", and closes any VC to that neighbor that has been asso- ciated with ATMARP_Sync traffic. The anti-entropy procedure for a neighbor terminates when all "not hot" ATMARP table entries have been successfully sent to that neighbor. The details on transitioning a neighbor from "down" to "up" status is presented later in this section. 3. After all anti-entropy processes have terminated, the ATMARP server may begin normal "hot" rumor mongering processing. 8.7.4 Rumor Mongering - The Sender The following guidelines govern the sending rumor mongering process: 1. Each table entry marked as "hot" is treated as a separate rumor. 2. For each "hot" rumor, the server maintains awareness of which of its neighbors it has infected with that rumor. If the "hot" rumor was received from an ATMARP_Sync ATMARP_Rumor message, the neighbor which sent the message is marked as infective before rumor mongering is started. 3. For each uninfected neighbor, the server selects a neighbor at random and transmits an uniquely serialized ATMARP_Sync rumor message and waits for an ATMARP_Sync acknowledgment for that serial number. If an ACK_Heard or ACK_NotHeard is received, the server marks the neighbor as "infected", examines the feedback, and notes whether the neighbor has heard the rumor: i.e., and ACK_Heard versus an ACK_NotHeard. If three or more neighbors have already heard the rumor or if all neighbors are infected, the table entry is marked as "not hot" and rumor mongering termi- nates for that entry. If an ATMARP_Sync acknowledgement is not received within 5 seconds, the sending server marks the neighbor as "down", and closes any VC to that neighbor that has been asso- ciated with ATMARP_Sync traffic. The details on transitioning a neighbor from "down" to "up" status is presented later in this Laubach [Page 24] DRAFT Classical IP and ARP over ATM Update August 1995 section. 8.7.5 Rumor Mongering - The Receiver The following guidelines apply to receiving a rumor. An ATMARP server, upon receiving an ATMARP_Sync ATMARP_Rumor message, examines its ATMARP table: 1. If an entry does not exist, a new entry is created based upon the rumor information presented in the message, it is marked as "hot", the entry's origin address is set to the ATMARP_Sync mes- sage's indicated origin address, and the the origin timestamp is set to the ATMARP_Sync message's indicated origin timestamp, and the entry local timestamp is initialized. An ATMARP_Sync ACK_NotHeard is returned, indicating the same serial number as the received ATMARP_Sync ATMARP_Rumor. 2. If an entry exists, the server examines the indicated origin information. If the message's origin address is the same as the table's origin address and the message's origin timestamp is the same as the table origin timestamp: the local timestamp is not updated; the entry is NOT marked as "hot"; and an ATMARP_Sync ACK_Heard is returned indicating the same serial number as the received ATMARP_Sync ATMARP_Rumor. 3. If an entry exists, the server examines the indicated origin information. If the message's origin address is the same as the table's origin address and the message's origin timestamp is NOT the same as the table origin timestamp: the origin timestamp is updated to the message's origin timestamp; the local timestamp is updated; the entry is marked as "hot"; and an ATMARP_Sync ACK_NotHeard is returned indicating the same serial number as the received ATMARP_Sync ATMARP_Rumor. 4. If an entry exists, the server examines the indicated origin information. If the message's origin address is NOT the same as the table's origin address: the origin address is updated to the message's origin timestamp; the origin timestamp is updated to the message's origin timestamp; the local timestamp is updated; the entry is marked as "hot"; and an ATMARP_Sync ACK_NotHeard is returned indicating the same serial number as the received ATMARP_Sync. 8.7.6 Restarting Connections to "Down" Neighbors The following guidelines apply to reopening and restarting an LLC/SNAP ATMARP_Sync message VC to a neighbor which has been previ- ously marked as "down". Laubach [Page 25] DRAFT Classical IP and ARP over ATM Update August 1995 1. An ATMARP server MUST detect if it has received a successful con- nection setup for an LLC/SNAP VC from a neighbor which has been previously marked as "down". If any ATMARP_Sync messages are received on that VC, the neighbor is then marked as "up", any local restart or waiting process is terminated, and an anti- entropy synchronization process is started with that neighbor (as described in Section 8.7.3 Server Startup Operation). Note that normal "hot" rumor mongering to that neighbor is suspended while any anti-entropy synchronization process is in progress. 2. An ATMARP server waits for a time period between 4.0 to 6.0 min- utes before attempting to restart a connection to a "down" neigh- bor. The actual time between 4.0 and 6.0 minutes is selected at random by the server. After the wait period expires, if the neighbor state is still "down", the server attempts to open an LLC/SNAP VC to that neighbor. After a successful opening of the VC, an anti-entropy synchronization process is started with that neighbor (as described in Section 8.7.3 Server Startup Opera- tion). After the first ATMARP_Sync acknowledgement (ACK_Heard or ACK_NotHeard) is received from the neighbor, that neighbor is marked as "up". Note that normal "hot" rumor mongering to any neighbor is suspended while the anti-entropy synchronization pro- cess is in progress. 8.7.7 Managing Multiple VCs Between Neighbors An ATMARP server MUST maintain only one synchronization VC between itself and a neighbor server. The following guidelines apply to closing VCs when multiple ATMARP synchronization VCs have been opened between the same two neighbors. 1. An ATMARP server, if it detects it has more than one synchroniza- tion VC open to a neighbor, MUST examine the serial numbers of the ATMARP_Sync messages that have been received and transmitted on the VCs in question. It is presumed that the "dueling" neigh- bors are each transmitting their ATMARP_Sync messages on the VCs that they opened. 2. The ATMARP server performs an integer comparison between the two serial numbers. If the serial numbers differ, the VC associated with the lesser serial number is closed and the remaining open VC is used for server-to-server messages. If the serial numbers are equal, the ATMARP server waits for 5 minutes before attempting the comparison again. 3. In the case of multiple synchronization VCs between an ATMARP server and a neighbor: if a server notes a remote initiated VC closure on one of the VCs, it MUST shift its use to one of the Laubach [Page 26] DRAFT Classical IP and ARP over ATM Update August 1995 remaining open VCs for ATM synchronization messages, and continue normal operation; i.e., if there are multiple synchronization VCs open between neighbors and one VC closes, this is normal, take note, and continue normal operation using a remaining open syn- chronization VC to that neighbor. 8.8 ATMARP Packet Formats Internet addresses are assigned independently of ATM addresses. Each host implementation MUST know its own IP and ATM address(es) and MUST respond to address resolution requests appropriately. IP members MUST also use ATMARP and InATMARP to resolve IP addresses to ATM addresses when needed. Note: the ATMARP packet format presented in this memo is general in nature in that the ATM number and ATM subaddress fields SHOULD map directly to the corresponding UNI 3.1 fields used for ATM call/connection setup signalling messages. The IP over ATM Work- ing Group expects ATM Forum NSAPA numbers (Structure 1) to pre- dominate over E.164 numbers (Structure 2) as ATM endpoint identi- fiers within ATM LANs. The ATM Forum's VC Routing specification is not complete at this time and therefore its impact on the operational use of ATM Address Structure 3 is undefined. The ATM Forum will be defining this relationship in the future. It is for this reason that IP members need to support all three ATM address structures. 8.8.1 ATMARP/InATMARP Request and Reply Packet Formats The ATMARP and InATMARP request and reply protocols use the same hardware type (ar$hrd), protocol type (ar$pro), and operation code (ar$op) data formats as the ARP and InARP protocols [3,12]. The location of these three fields within the ATMARP packet are in the same byte position as those in ARP and InARP packets. A unique hardware type value has been assigned for ATMARP. In addition, ATMARP makes use of an additional operation code for ARP_NAK. The remainder of the ATMARP/InATMARP packet format is different than the ARP/InARP packet format. The ATMARP and InATMARP protocols have several fields that have the following format and values: Laubach [Page 27] DRAFT Classical IP and ARP over ATM Update August 1995 Data: ar$hrd 16 bits Hardware type ar$pro 16 bits Protocol type ar$shtl 8 bits Type & length (TL) of source ATM number (q) ar$sstl 8 bits Type & length (TL) of source ATM subaddress (r) ar$op 16 bits Operation code (request, reply, or NAK) ar$spln 8 bits Length of source protocol address (s) ar$thtl 8 bits Type & length (TL) of target ATM number (x) ar$tstl 8 bits Type & length (TL) of target ATM subaddress (y) ar$tpln 8 bits Length of target protocol address (z) ar$sha qoctets of source ATM number ar$ssa roctets of source ATM subaddress ar$spa soctets of source protocol address ar$tha xoctets of target ATM number ar$tsa yoctets of target ATM subaddress ar$tpa zoctets of target protocol address Where: ar$hrd - assigned to ATM Forum address family and is 19 decimal (0x0013) [4]. ar$pro - see Assigned Numbers for protocol type number for the protocol using ATMARP. (IP is 0x0800). [changebar on] ar$shtl - Type and length of source ATM number. See Section 8.8.4 for TL encoding details. ar$sstl - Type and length of source ATM subaddress. See Section 8.8.4 for TL encoding details. ar$op - The operation type value (decimal): ATMARP_Request = ARP_REQUEST = 1 ATMARP_Reply = ARP_REPLY = 2 InATMARP_Request = InARP_REQUEST = 8 InATMARP_Reply = InARP_REPLY = 9 ATMARP_NAK = ARP_NAK = 10 ar$spln - length in octets of the source protocol address. Value range is 0 or 4 (decimal). For IPv4 ar$spln is 4. ar$thtl - Type and length of target ATM number. See Section 8.8.4 for TL encoding details. ar$tstl - Type and length of target ATM subaddress. See Section 8.8.4 for TL encoding details. Laubach [Page 28] DRAFT Classical IP and ARP over ATM Update August 1995 [changebar off] ar$tpln - length in octets of the target protocol address. Value range is 0 or 4 (decimal). For IPv4 ar$tpln is 4. ar$sha - source ATM number (E.164 or ATM Forum NSAPA) ar$ssa - source ATM subaddress (ATM Forum NSAPA) ar$spa - source protocol address ar$tha - target ATM number (E.164 or ATM Forum NSAPA) ar$tsa - target ATM subaddress (ATM Forum NSAPA) ar$tpa - target protocol address 8.8.2 ATMARP Synchronization Packet Format The ATMARP Synchronization protocol packet uses the same hardware type (ar$hrd), protocol type (ar$pro), and operation code (ar$op) data formats as the ATMARP and InATMARP packets. The location of these three fields within the ATMARP packet are in the same byte position as those in ARP and InARP packets [2,3]. The ATMARP_Sync message makes use of an additional operation code (ar$op) for ATMARP_Sync. The type of synchronization packet is specified in the Sync message type field (ar$styp). This memo defines four synchro- nization message types: ATMARP rumor synchronization (ATMARP_Rumor); acknowledgement and the rumor has not been previously heard (ACK_NotHeard); acknowledgement and the rumor has been heard previ- ously (ACK_Heard); and the requested synchronization type is not sup- ported. ATMARP rumor synchronization packets use an additional message field (ar$msg) to encode the ATMARP rumor that is being passed. This mes- sage field is not present on ACK_NotHeard or ACK_Heard messages. Note: it is the intent of this memo to acknowledge the rumoring mech- anism is general in nature and that future ATMARP synchronization message types can be defined based on this packet structure such that the message field (ar$msg) is exploited for new synchronization mes- sage types. Laubach [Page 29] DRAFT Classical IP and ARP over ATM Update August 1995 Data: ar$hrd 16 bits Hardware type ar$pro 16 bits Protocol type ar$ohtl 8 bits Type & length (TL) of origin ATM number (q) ar$ostl 8 bits Type & length (TL) of origin ATM subaddress (r) ar$op 16 bits Operation code (ATMARP_Sync) ar$sn 32 bits Serial Number ar$ots 32 bits Origin timestamp ar$styp 8 bits Sync message type (ACK_NotHeard/Heard, ATMARP_Rumor) ar$oha qoctets of origin ATM number ar$osa roctets of origin ATM subaddress ar$msg moctets of synchronization message Where: ar$hrd - assigned to ATM Forum address family and is 19 decimal (0x0013) [4]. ar$pro - see Assigned Numbers for protocol type number for the protocol using ATMARP. (IP is 0x0800). ar$ohtl - Type and length of origin server's ATM number. See Section 8.8.4 for TL encoding details. ar$ostl - Type and length of origin server's ATM subaddress. See Section 8.8.4 for TL encoding details. ar$op - The operation type value (decimal): ATMARP_Sync = ?? ar$osn - 32 bit unsigned integer representing the sending server's serial number for this ATMARP synchronization packet. ar$ots - 32 bit unsigned integer representing the origin timestamp of the originating server for this message. ar$styp - 8 bit (1 octet) unsigned integer representing the ATMARP synchronization message type. The value in decimal for the messages are: SYNC_Unsup = 0 ATMARP_Rumor = 1 ACK_NotHeard = 2 ACK_Heard = 3 ar$oha - origin ATM number (E.164 or ATM Forum NSAPA). This Laubach [Page 30] DRAFT Classical IP and ARP over ATM Update August 1995 is the ATM number of the originating server. ar$osa - origin ATM subaddress (ATM Forum NSAPA). If present, this is the ATM subaddress of the originating server. ar$msg - synchronization message For ATMARP_Rumor synchronization messages, ar$msg is variable in length and encoded with the following values: Data: ar$rhtl 8 bits Type & length (TL) of rumor ATM number (x) ar$rstl 8 bits Type & length (TL) of rumor ATM subaddress (y) ar$rpln 8 bits Length of rumor protocol address (z) ar$rha xoctets of rumor ATM number ar$rsa yoctets of rumor ATM subaddress ar$rpa zoctets of rumor protocol address Where: ar$rhtl - Type and length of rumor ATM number. See Section 8.8.4 for TL encoding details. ar$rstl - Type and length of rumor ATM subaddress. See Section 8.8.4 for TL encoding details. ar$rpln - Length in octets of the rumor protocol address. For IPv4 ar$rpln is 4. ar$rha - Rumor ATM number (E.164 or ATM Forum NSAPA) ar$rsa - Rumor ATM subaddress (ATM Forum NSAPA) ar$rpa - Rumor protocol address The coding of these values in ar$msg obey the same conventions as presented in the ATMARP and InATMARP packet format. The specifics on coding TL and ATM addresses is detailed in Section 8.8.4. The rumored information content for these fields is exactly replicated from the ATMARP client registration information presented to the originating server. For ACK_Heard and ACK_NotHeard message types, the ar$msg field is not used and no storage is allocated for the field in the packet. For these message types, the received ATMARP_Sync ATMARP_Rumor packet is kept in its entirety. When the station is ready to transmit an acknowledgement, the ATMARP_Sync packet data is exactly copied (e.g. using bcopy) for transmission omitting the trailing ar$msg field (ar$msg storage is not allocated) and the synchronization message Laubach [Page 31] DRAFT Classical IP and ARP over ATM Update August 1995 type code (ar$styp) is changed from ATMARP_Rumor to ACK_NotHeard or ACK_Heard as appropriate for the feedback from the receiving ATMARP server. Receiving Unknown Synchronization Message Types If an ATMARP server receives an ATMARP_Sync message type (ar$styp) for which it is not coded to support, the server must transmit a Syn- chronization Type Unsupported (SYNC_Unsup) message to the sender using the same VC on which the unsupported ATMARP Synchronization message was received. The unsupported message is constructed by 1) exactly copying the ATMARP_Sync ATMARP_Rumor packet data (e.g. using bcopy); 2) replacing the trailing ar$msg field (if present) with the unsupported return message; and 3) changing the synchronization mes- sage type code (ar$styp) to SYNC_Unsup. The new message field (ar$msg) is constructed as a 1 octet counter followed by an array of 1 octet ar$styp values, with one array ele- ment set to each of the synchronization message types which are sup- ported by this server. For servers coded to this memo, the array returned then encodes ATMARP_Rumor, ACK_NotHeard, and ACK_Heard. which occupies 3 octets of array message. Synchronization message types MUST be listed in ascending numerical order and each type MUST only be listed once in the array. The SYNC_Unsup type is implied with receipt of this message. For SYNC_Unsup messages, ar$msg is variable in length and encoded with the following values: Data: ar$ucnt 1 octet Counter of elements in the array (u) ar$uary uoctets Array of unsupported message types Where: ar$ucnt - 1 Octet unsigned integer representing the count of the number of elements in the ar$uary array. ar$uary - Array of 1 octet ar$styp values representing the synchronization message types this server supports. For servers implemented to this memo, ar$msg is 4 octets in length and is coded as follows: Laubach [Page 32] DRAFT Classical IP and ARP over ATM Update August 1995 1 octet Unsigned Integer (Hex) Message Type MSB LSB +-------------+ Counter | 0x03 | +-------------+ Element 0 | 0x01 | ATMARP_Rumor +-------------+ 1 | 0x02 | ACK_NotHeard +-------------+ 2 | 0x03 | ACK_Heard +-------------+ This protocol feature allows future extensibility of the synchroniza- tion protocol for new feature addition while maintaining backwards compatibility for then-existing implementations. 8.8.3 Receiving Unknown ATMARP packets If an ATMARP client receives an ATMARP message with an operation code (ar$op) for which it is not coded to support, it MUST gracefully dis- card the message and continue normal operation. An ATMARP client is NOT REQUIRED to return any message to the sender of the unsupported message. If an ATMARP server receives an ATMARP message with an operation code (ar$op) for which it is not coded to support, the server MUST return an ATMARP Opcode Not Supported (ATMARP_Unsup) message to the sender using the same VC on which the unsupported ATMARP message was received. The format of the message is as follows: Data: ar$hrd 16 bits Hardware type ar$pro 16 bits Protocol type ar$shtl 8 bits Type & length (TL) of server ATM number (q) ar$sstl 8 bits Type & length (TL) of server ATM subaddress (r) ar$op 16 bits Operation code (ATMARP_Unsup) ar$sha qoctets of server ATM number ar$ssa roctets of server ATM subaddress ar$opcnt 8 bits Number of entries in ar$opary - opcode count (w) ar$opary w*16 bits Array of w * ar$op values Laubach [Page 33] DRAFT Classical IP and ARP over ATM Update August 1995 Where: ar$hrd - assigned to ATM Forum address family and is 19 decimal (0x0013) [4]. ar$pro - see Assigned Numbers for protocol type number for the protocol using ATMARP. (IP is 0x0800). ar$shtl - Type and length of server ATM number. This is the ATM number of the server sending this message. See Section 8.8.4 for TL encoding details. ar$sstl - Type and length of client ATM subaddress. This is the ATM subaddress of the server sending this message. See Section 8.8.4 for TL encoding details. ar$op - The operation type value (decimal): ATMARP_Unsup = ?? ar$sha - server ATM number (E.164 or ATM Forum NSAPA) ar$ssa - server ATM subaddress (ATM Forum NSAPA) ar$opcnt - 8 bit unsigned integer representing the number of entries in ar$opary. ar$opary - Array of 16-bit ar$op values representing the ATMARP operation codes this server supports. The ar$opary element is constructed as an array of unsigned 16-bit ar$op values, with one array element set to each of the ATMARP opera- tion codes which are supported by this server. For servers coded to this memo, the array returned encodes the ATMARP_Request, ATMARP_Reply, InATMARP_Request, InATMARP_Reply, ATMARP_Nak, and ATMARP_Sync opcodes. Opcodes MUST be listed in ascending numerical order. Each opcode MUST only be listed once in the array. The ATMARP_Unsup opcode is implied with receipt of this message. For servers implemented to this memo, ar$opcnt has a value of 6 deci- mal and ar$opary array coded as follows: Laubach [Page 34] DRAFT Classical IP and ARP over ATM Update August 1995 16-bit Unsigned Integer (Hex) Message Type MSB LSB +---------------+ Element 0 | 0x0001 | ATMARP_Request +---------------+ 1 | 0x0002 | ATMARP_Reply +---------------+ 2 | 0x0008 | InATMARP_Request +---------------+ 3 | 0x0009 | InATMARP_Reply +---------------+ 4 | 0x000A | ATMARP_Nak +---------------+ 5 | 0x00?? | ATMARP_Sync +---------------+ If an ATMARP client receives an ATMARP message with an operation code (ar$op) of ATMARP_Unsup, it MUST gracefully process the message and continue normal operation. The client MAY wish to take other imple- mentation dependent action based upon the contents of this message. Any such action is beyond the scope of this memo. This protocol message allows future extensibility of the ATMARP pro- tocol for new feature additions of ATMARP servers while maintaining backwards compatibility for then-existing implementations. 8.8.4 TL, ATM Number, and ATM Subaddress Encoding The encoding of the 8-bit TL (type and length) fields in ATMARP and In_ATMARP packets is as follows: MSB 8 7 6 5 4 3 2 1 LSB +-----+-----+-----+-----+-----+-----+-----+-----+ | 0 | 1/0 | Octet length of address | +-----+-----+-----+-----+-----+-----+-----+-----+ Where: bit.8 (reserved) = 0 (for future use) bit.7 (type) = 0 ATM Forum NSAPA format = 1 E.164 format bit.6-1 (length) = 6 bit unsigned octet length of address (MSB = bit.6, LSB = bit.1) Value range is from 0 to 20 (decimal). Laubach [Page 35] DRAFT Classical IP and ARP over ATM Update August 1995 [changebar on] ATM addresses, as defined by the ATM Forum UNI 3.1 signaling specifi- cation [9], include a "Calling Party Number Information Element" and a "Calling Party Subaddress Information Element". These Information Elements (IEs) SHOULD map to ATMARP/InATMARP source ATM number and source ATM subaddress respectively. Furthermore, ATM Forum defines a "Called Party Number Information Element" and a "Called Party Subad- dress Information Element". These IEs map to ATMARP/InATMARP target ATM number and target ATM subaddress respectively. [changebar off] The ATM Forum defines three structures for the combined use of number and subaddress [9]: ATM Number ATM Subaddress -------------- -------------- Structure 1 ATM Forum NSAPA null Structure 2 E.164 null Structure 3 E.164 ATM Forum NSAPA [changebar on] ATMARP and InATMARP requests and replies for ATM address structures 1 and 2 MUST indicate a null or unknown ATM subaddress by setting the appropriate subaddress length to zero; i.e. ar$sstl.length = 0 or ar$tstl.length = 0, the corresponding type field (ar$sstl.type or ar$tstl.type) MUST be ignored and the physical space for the ATM sub- address buffer MUST not be allocated in the ATMARP packet. For exam- ple, if ar$sstl.length=0, the storage for the source ATM subaddress is not allocated and the first byte of the source protocol address ar$spa follows immediately after the last byte of the source hardware address ar$sha in the packet. Null or unknown ATM addresses are MUST be indicated by setting the appropriate address length to zero; i.e., ar$shtl.length and ar$thtl.length is zero and the corresponding type field (ar$sstl.type or ar$tstl.type) MUST be ignored and the physical space for the ATM address or ATM subaddress buffer MUST not be allocated in the ATMARP packet. 8.8.5 ATMARP_NAK Packet Format The ATMARP_NAK packet format is the same as the received ATMARP_Request packet format with the operation code set to ARP_NAK, i.e., the ATMARP_Request packet data is exactly copied (e.g. using bcopy) for transmission with the ATMARP_Request operation code Laubach [Page 36] DRAFT Classical IP and ARP over ATM Update August 1995 changed to ARP_NAK value. 8.8.6 Variable Length Requirements for ATMARP Packets ATMARP, InATMARP, and ATMARP_Sync packets are variable in length. A null or unknown source or target protocol address is indicated by the corresponding length set to zero: e.g., when ar$spln or ar$tpln is zero the physical space for the corresponding address structure MUST not be allocated in the packet. For backward compatibility with previous implementations, a null IPv4 protocol address may be received with length = 4 and an allocated address in storage set to the value 0.0.0.0. Receiving stations MUST be liberal in accepting this format of a null IPv4 address, however on transmitting an ATMARP or InATMARP packet, a null IPv4 address MUST only be indicated by the length set to zero and MUST have no storage allocated. [changebar off] 8.8 ATMARP/InATMARP Packet Encapsulation ATMARP and InATMARP packets are to be encoded in AAL5 PDUs using LLC/SNAP encapsulation. The format of the AAL5 CPCS-SDU payload field for ATMARP/InATMARP PDUs is: Payload Format for ATMARP/InATMARP PDUs: +------------------------------+ | LLC 0xAA-AA-03 | +------------------------------+ | OUI 0x00-00-00 | +------------------------------+ | Ethertype 0x08-06 | +------------------------------+ | | | ATMARP/InATMARP Packet | | | +------------------------------+ The LLC value of 0xAA-AA-03 (3 octets) indicates the presence of a SNAP header. The OUI value of 0x00-00-00 (3 octets) indicates that the following two-bytes is an ethertype. The Ethertype value of 0x08-06 (2 octets) indicates ARP [4]. Laubach [Page 37] DRAFT Classical IP and ARP over ATM Update August 1995 The total size of the LLC/SNAP header is fixed at 8-octets. This aligns the start of the ATMARP packet on a 64-bit boundary relative to the start of the AAL5 CPCS-SDU. The LLC/SNAP encapsulation for ATMARP/InATMARP presented here is con- sistent with the treatment of multiprotocol encapsulation of IP over ATM AAL5 as specified in [2] and in the format of ATMARP over IEEE 802 networks as specified in [5]. Traditionally, address resolution requests are broadcast to all directly connected IP members within a LIS. It is conceivable in the future that larger scaled ATM networks may handle ATMARP requests to destinations outside the originating LIS, perhaps even globally; issues raised by ATMARPing outside the LIS or by a global ATMARP mechanism are beyond the scope of this memo. 9. IP Broadcast Address ATM does not support broadcast addressing, therefore there are no mappings available from IP broadcast addresses to ATM broadcast ser- vices. Note: this lack of mapping does not restrict members from transmitting or receiving IP datagrams specifying any of the four standard IP broadcast address forms as described in [8]. Members, upon receiving an IP broadcast or IP subnet broadcast for their LIS, MUST process the packet as if addressed to that station. [changebar on] This memo recognizes the future development of standards and imple- mentations that will extend the operations as defined in this memo to provide an IP broadcast capability for use by the classical client. [changebar off] 10. IP Multicast Address ATM does not support multicast address services, therefore there are no mappings available from IP multicast addresses to ATM multicast services. Current IP multicast implementations (i.e., MBONE and IP tunneling, see [10]) will continue to operate over ATM based logical IP subnets if operated in the WAN configuration. This memo recognizes the future development of ATM multicast service addressing by the ATM Forum. When available and widely implemented, the roll-over from the current IP multicast architecture to this new ATM architecture will be straightforward. [changebar on] Laubach [Page 38] DRAFT Classical IP and ARP over ATM Update August 1995 This memo recognizes the future development of standards and imple- mentations that will extend the operations as defined in this memo to provide an IP multicast capability for use by the classical client. [changebar off] 11. Security Not all of the security issues relating to IP over ATM are clearly understood at this time, due to the fluid state of ATM specifica- tions, newness of the technology, and other factors. It is believed that ATM and IP facilities for authenticated call man- agement, authenticated end-to-end communications, and data encryption will be needed in globally connected ATM networks. Such future secu- rity facilities and their use by IP networks are beyond the scope of this memo. There are known security issues relating to host impersonation via the address resolution protocols used in the Internet [13]. No spe- cial security mechanisms have been added to the address resolution mechanism defined here for use with networks using IP over ATM. 11. MIB Specification This section to be completed in a later revision of the Internet- Draft. 12. Open Issues o Automatic configuration of client ATM addresses via DHCP [19] or via ATM UNI 3.1 Interim Local Management Interface (ILMI) ser- vices would be a useful extended service addition to this docu- ment and should be addressed in a separate memo. o ATMARP packets are not authenticated. This is a potentially serious flaw in the overall system by allowing a mechanism by which corrupt information may be introduced into the server sys- tem. o ATMARP_Sync rumor messages are currently unauthenticated. It is possible to extend the protocol by adding an authenticated rumor type and specifying the new message details and operational requirements. Laubach [Page 39] DRAFT Classical IP and ARP over ATM Update August 1995 REFERENCES [1] Piscitello, D., and Lawrence, J., "IP and ARP over the SMDS Ser- vice", RFC-1209, USC/Information Sciences Institute, March 1991. [2] Heinanen, Juha, "Multiprotocol Encapsulation over ATM Adaptation Layer 5", RFC-1483, USC/Information Sciences Institute, July 1993. [3] Plummer, D., "An Ethernet Address Resolution Protocol - or - Con- verting Network Addresses to 48.bit Ethernet Address for Trans- mission on Ethernet Hardware", RFC-826, MIT, November 1982. [4] Reynolds, J., and Postel, J., "Assigned Numbers", RFC-1340, USC/ Information Sciences Institute, July 1992. [5] Postel, J., and Reynolds, J., "A Standard for the Transmission of IP Datagrams over IEEE 802 Networks", RFC-1042, USC/Information Sciences Institute, February 1988. [6] CCITT, "Draft Recommendation I.363", CCITT Study Group XVIII, Geneva, 19-29 January 1993. [7] CCITT, "Draft text for Q.93B", CCITT Study Group XI, 23 September - 2 October 1992. [8] Braden, R., "Requirements for Internet Hosts -- Communication Layers", RFC-1122, USC/Information Sciences Institute, October 1989. [9] ATM Forum, "ATM User-Network Interface (UNI) Specification Ver- sion 3.1.", ISBN 0-13-393828-X, Prentice-Hall, Inc., Upper Saddle River, NJ, 07458, September, 1994. [10] Deering, S, "Host Extensions for IP Multicasting", RFC-1112, USC/Information Sciences Institute, August 1989. [11] Colella, Richard, and Gardner, Ella, and Callon, Ross, "Guide- lines for OSI NSAP Allocation in the Internet", RFC-1237, USC/Information Sciences Institute, July 1991. [12] Bradely, T., and Brown, C., "Inverse Address Resolution Proto- col", RFC-1293, USC/Information Sciences Institute, January 1992. [13] Bellovin, Steven M., "Security Problems in the TCP/IP Protocol Suite", ACM Computer Communications Review, Vol. 19, Issue 2, pp. 32-48, 1989. Laubach [Page 40] DRAFT Classical IP and ARP over ATM Update August 1995 [14] Demers, A., Gealy, M., Greene, D., Hauser, C., Irish, W., Lar- son, J., Manning, S., Shenker, S., Sturgis, H., Swinehart, D., Terry, D., and Woods, D., "Epidemic Algorithms for Replicated Database Maintenance", CSL-89-1, Xerox Corporation, January, 1989. [15] Knowles, S., "IESG Advice from Experience with Path MTU Discov- ery, RFC-1435, IESG, March 1993. [16] Kent C., and J. Mogul, "Fragmentation Considered Harmful", Pro- ceedings of the ACM SIGCOMM '87 Workshop on Frontiers in Computer Communications Technology, August 1987. [17] Mogul, J., and S. Deering, "Path MTU Discovery", RFC-1191, DECWRL, Stanford University, November 1990. [18] Katz, D., and Piscitello, D., "NBMA Next Hope Resolution Proto- col", draft-ietf-rolc-nhrp-04.txt, Internet Draft (work in progress), May 1995. [19] Droms, R., "Dynamic Host Configuration Protocol", RFC1541, Buck- nell University, October 1993. [20] Bound, J., Carpenter, B., Harrington, J., Houldsworth, J., and Lloyd, A., "OSI NSAPs and IPv6", draft-ietf-ipngwg- Internet Draft, (work in progress proposed for elevation to experimental status), August 1995. [21] Hinden, R, and Deering, S. "IP Version 6 Addressing Architec- ture", draft-ietf-ipngwg-addr-arch-03.txt, Internet Draft (work in progress), June 1995. Author's Address Mark Laubach Com21, Inc. 1991 Landings Drive Mountain View, CA 94043 Phone: 415.254.5882 FAX: 415.254.5883 EMail: laubach@com21.com Laubach [Page 41] DRAFT Classical IP and ARP over ATM Update August 1995 Appendix A. Exploiting IPv4 Addresses inside an NSAPA in a LIS The procedures presented in this appendix MUST not change the implementation or interfere with the normal operation of any IP over ATM client. If implemented, the address format presented in this section MUST be followed, otherwise the information presented in this appendix in strictly informational in nature and should be regarded as suggested guidelines should such a procedure be used within an LIS. A.1 General This appendix describes a method where an IPv4 address could be automatically embedded inside an NSAPA using an automatic procedure within the ATMARP server. This would allow an LIS to operate using IPv4 addresses as ATM end point addresses without impacting client implementation; i.e., the entire process is carried out in a modified ATMARP server. The general requirements are that the ATM network for the LIS be capable of using this embedded NSAPA format. The ATMARP server would be modified so that it knows the range of IPv4 addresses over which the algorithmic translation would be applied. If the target protocol address were within the appropriate ranges, the translation would be applied to the target protocol of any ATMARP Request to produce the target ATM number for the ATMARP Reply. ATM subaddresses would not be used in this model. The client, after receiving the ATMARP Reply, would attempt to open a connection to the ATM endpoint address specified in the ATMARP Reply. If the ATMARP Request target protocol address was outside the range of IPv4 addresses, the ATMARP server should return an ATMARP NAK to the client. A.2 IPv4/NSAPA Embedded Format If it is required, for whatever reason, to embed an IPv4 address inside a 20-octet NSAP address, then the following format MUST be used. The specific use of this embedding is to express an IPv4 address within the ATM Forum address format. The IPv4 address is actually embedded first in a 16-byte IPv6 NSAPA address using the conventions being developed in the IETF IPNG Working Group [20,21]. The first three octets are an IDP in binary ICD format, using the ICD code allocated to the IANA, and meaning "this NSAPA embeds a 16 byte IPv6 address". The last octet remains the NSAPA selector field. Laubach [Page 42] DRAFT Classical IP and ARP over ATM Update August 1995 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 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 0-3 | AFI = 47 | ICD = 0090 | 0x00 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 4-7 | 0x00000000 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 8-11 | 0x00000000 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 12-15| 0x000000 | IPv4 (byte 0) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 16-19| IPv4 (bytes 1-3) | 0x00 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ B.3 Impact on Server-to-Server Synchronization If anyone wishes to implement this convention for all hosts located within a LIS, the ATMARP server synchronization mechanisms need not be used; the entire ATMARP Reply message is algorithmically derived from the ATMARP Request information, hence no ATMARP table informa- tion is needed within an ATMARP server, and no information need be shared with other servers. The ATMARP server selection guidelines could still be followed as there may be more than one adapted ATMARP server providing the address resolution service for the LIS. So long as all ATMARP servers are configured with the same range of IPv4 addresses, the size of the LIS could be arbitrarily large, and each client would have access to more than one server for ATMARP address resolution. By supporting the use of ATMARP NAK messages, NHRP servers could still be used to augment direct ATM connectivity with hosts located outside the LIS. Laubach [Page 43]