Request for Comments: 2844
OSPF over ATM and Proxy-PAR
Status of this Memo
This memo defines an Experimental Protocol for the Internet community. It does not specify an Internet standard of any kind. Discussion and suggestions for improvement are requested. Distribution of this memo is unlimited.
Copyright © The Internet Society (2000). All Rights Reserved.
This memo specifies, for OSPF implementors and users, mechanisms describing how the protocol operates in ATM networks over PVC and SVC meshes with the presence of Proxy-PAR. These recommendations require no protocol changes and allow simpler, more efficient and cost- effective network designs. It is recommended that OSPF implementations should be able to support logical interfaces, each consisting of one or more virtual circuits and used either as numbered logical point-to-point links (one VC), logical NBMA networks (more than one VC) or Point-to-MultiPoint networks (more than one VC), where a solution simulating broadcast interfaces is not appropriate. PAR can help distribute across the ATM cloud configuration setup and changes of such interfaces when OSPF capable routers are (re-)configured. Proxy-PAR can in turn be used to exchange this information between the ATM cloud and the routers connected to it.
Proxy-PAR and PAR have been accepted as standards by the ATM Forum in January 1999 . A more complete overview of Proxy-PAR than in the section below is given in .
1.1 Introduction to Proxy-PAR
Proxy-PAR  is an extension that allows different ATM attached devices (like routers) to interact with PAR-capable switches and to query information about non-ATM services without executing PAR themselves. The Proxy-PAR client side in the ATM attached device is much simpler in terms of implementation complexity and memory requirements than a complete PAR protocol stack (which includes the full PNNI  protocol stack) and should allow easy implementation, e.g. in existing IP routers. In addition, clients can use Proxy-PAR to register the various non-ATM services and protocols they support. Proxy PAR has consciously been omitted as part of ILMI  due to the complexity of PAR information passed in the protocol and the fact that it is intended for integration of non-ATM protocols and services only. A device that executes Proxy-PAR does not necessarily need to execute ILMI or UNI signaling, although this normally will be the case.
The protocol in itself does not specify how the distributed service registration and data delivered to the client is supposed to drive other protocols. Hence OSPF routers, for instance, that find themselves through Proxy-PAR could use this information in a Classical IP and ARP over ATM  fashion, forming a full mesh of point-to-point connections to interact with each other to simulate broadcast interfaces. For the same purpose, LANE  or MARS  could be used. As a byproduct, Proxy-PAR could provide the ATM address resolution for IP-attached devices, but such resolution can be achieved by other protocols under specification at the IETF as well, e.g. . Last but not least, it should be mentioned here that the protocol coexists with and complements the ongoing work in IETF on server detection via ILMI extensions [9,10,11].
1.1.1 Proxy-PAR Scopes
Any information registered through Proxy-PAR is flooded only within a defined scope that is established during registration and is equivalent to the PNNI routing level. As no assumption can be made about the information distributed (e.g. IP addresses bound to NSAPs are not assumed to be aligned with them in any respect such as encapsulation or functional mapping), it cannot be summarized. This makes a careful handling of scopes necessary to preserve the scalability. More details on the usage of scope can be found in .
1.2 Introduction to OSPF
OSPF (Open Shortest Path First) is an Interior Gateway Protocol (IGP) and described in  from which most of the following paragraphs has been taken almost literally. OSPF distributes routing information between routers belonging to a single Autonomous System. The OSPF protocol is based on link-state or SPF technology. It was developed by the OSPF working group of the Internet Engineering Task Force. It has been designed expressly for the TCP/IP internet environment, including explicit support for IP subnetting, and the tagging of externally-derived routing information. OSPF also utilizes IP multicast when sending/receiving the updates. In addition, much work has been done to produce a protocol that responds quickly to topology changes, yet involves small amounts of routing protocol traffic.
To cope with the needs of NBMA and demand-circuit-capable networks such as Frame Relay or X.25,  has been made available. It standardizes extensions to the protocol that allow efficient operation over on-demand circuits.
OSPF supports three types of networks today:
- Point-to-point networks: A network that joins a single pair of routers. Point-to-point networks can either be numbered or unnumbered. In the latter case the interfaces do not have IP addresses nor masks. Even when numbered, both sides of the link do not have to agree on the IP subnet.
- Broadcast networks: Networks supporting many (more than two) attached routers, together with the capability of addressing a single physical message to all of the attached routers (broadcast). Neighboring routers are discovered dynamically on these networks using the OSPF Hello Protocol. The Hello Protocol itself takes advantage of the broadcast capability. The protocol makes further use of multicast capabilities, if they exist. An Ethernet is an example of a broadcast network.
+ Non-broadcast networks: Networks supporting many (more than two) attached routers, but having no broadcast capability. Neighboring routers are maintained on these nets using OSPF's Hello Protocol. However, due to the lack of broadcast capability, some configuration information is necessary for the correct operation of the Hello Protocol. On these networks, OSPF protocol packets that are normally multicast need to be sent to each neighboring router, in turn. An X.25 Public Data Network (PDN) is an example of a non-broadcast network.
OSPF runs in one of two modes over non-broadcast networks. The first mode, called non-broadcast multi-access (NBMA), simulates the operation of OSPF on a broadcast network. The second mode, called Point-to-MultiPoint, treats the non-broadcast network as a collection of point-to-point links. Non-broadcast networks are referred to as NBMA networks or Point-to-MultiPoint networks, depending on OSPF's mode of operation over the network.
2 OSPF over ATM
Contrary to broadcast-simulation-based solutions such as LANE  or Classical IP and ARP over ATM , this document elaborates on how to handle virtual OSPF interfaces over ATM such as NBMA, Point-to- MultiPoint or point-to-point and allow for their auto-configuration in the presence of Proxy-PAR. One advantage is the circumvention of server solutions that often present single points of failure or hold large amounts of configuration information.
The other main benefit is the capability of executing OSPF on top of NBMA and Point-to-MultiPoint ATM networks, and still benefit from the automatic discovery of OSPF neighbors. As opposed to broadcast networks, broadcast-simulation-based networks (such as LANE or Classical IP and ARP over ATM), and point-to-point networks, where an OSPF router dynamically discovers its neighbors by sending Hello packets to the All-SPFRouters multicast address, this is not the case on NBMA and Point-to-MultiPoint networks. On NBMA networks, the list of all other attached routers to the same NBMA network has to be manually configured or discovered by some other means: Proxy-PAR allows this configuration to be automated. Also on Point-to- MultiPoint networks, the set of routers that are directly reachable can either be manually configured or dynamically discovered by Proxy-PAR or mechanisms such as Inverse ATMARP. In an ATM network, (see 8.2 in ) Inverse ATMARP can be used to discover the IP address of the router at the remote end of a given PVC, whether or not its ATM address is known. But Inverse ATMARP does not return, for instance, whether the remote router is running OSPF, unlike Proxy- PAR.
Parallel to , which describes the recommended operation of OSPF over Frame Relay networks, a similar model is assumed where the underlying ATM network can be used to model single VCs as point-to- point interfaces or collections of VCs as non-broadcast interfaces, whether in NBMA or Point-to-MultiPoint mode. Such a VC or collection of VCs is called a logical interface and specified through its type (either point-to-point, NBMA or Point-to-MultiPoint), VPN ID (the
Virtual Private Network to which the interface belongs), address and mask. Layer 2 specific configurations such as the address resolution method, class and quality of service of circuits used, and others, must also be included. As a logical consequence thereof, a single, physical interface could encompass multiple IP subnets or even multiple VPNs. Contrary to layer 2 and IP addressing information, when running Proxy-PAR, most of the OSPF information needed to operate such a logical interface does not have to be configured into routers statically but can be provided through Proxy-PAR queries. This allows much more dynamic configuration of VC meshes in OSPF environments than, for example, Frame Relay solutions do.
Proxy-PAR queries can also be issued with a subnet address set to 0.0.0.0, instead of a specific subnet address. This type of query returns information on all OSPF routers available in all subnets within the scope specified in the query. This can be used for instance when the IP addressing information has not been configured.
2.2 Configuration of OSPF interfaces with Proxy-PAR
To achieve the goal of simplification of VC mesh reconfiguration, Proxy-PAR allows the router to learn automatically most of the configuration that has to be provided to OSPF. Non-broadcast and point-to-point interface information can be learned across an ATM cloud as described in the ongoing sections. It is up to the implementation to possibly allow for a mixture of Proxy-PAR autoconfiguration and manual configuration of neighbor information. Moreover, manual configuration could, for instance, override or complement information derived from a Proxy-PAR client. In addition, OSPF extensions to handle on-demand circuits  can be used to allow the graceful tearing down of VCs not carrying any OSPF traffic over prolonged periods of time. The various interactions are described in sections 2.2.1, 2.2.2 and 2.2.3.
Even after autoconfiguration of interfaces has been provided, the problem of VC setups in an ATM network is unsolved because none of the normally used mechanisms such as Classical IP and ARP over ATM  or LANE  are assumed to be present. Section 2.5 describes the behavior of OSPF routers necessary to allow for router connectivity.
2.2.1 Autoconfiguration of Non-Broadcast Multiple-Access (NMBA)
Proxy-PAR allows the autoconfiguation of the list of all routers residing on the same IP network in the same VPN by simply querying the Proxy-PAR server. Each router can easily obtain the list of all OSPF routers on the same subnet with their router priorities and corresponding ATM addresses. This is the precondition for OSPF to work properly across such logical NBMA interfaces. Note that this member list, when learned through Proxy-PAR queries, can dynamically change with PNNI (in)stability and general ATM network behavior. Relying on an OSPF mechanism to discover a lack of reachability in the overlaying logical IP network could alleviate the risk of thrashing DR elections and excessive information flooding. Once the DR election has been completed and the router has not been elected DR or BDR, an implementation of  can ignore the fact that all routers on the specific NBMA subnet are available in its configuration because it only needs to maintain VCs to the DR and BDR. Note that this information can serve other purposes, such as the forwarding of data packets (see section 2.4).
Traditionally, router configuration for a NBMA network provides the list of all neighboring routers to allow for proper protocol operation. For stability purposes, the user may choose to provide a list of neighbors through such static means but also enable the operation of Proxy-PAR protocol to complete the list. It is left up to specific router implementations to determine whether to use the manual configuration in addition to the information provided by Proxy-PAR, to use the manual configuration to filter dynamic information, or whether a concurrent mode of operation is prohibited. In any case it should be obvious that allowing for more flexibility may facilitate operation but provides more possibilities for misconfiguration as well.
2.2.2 Autoconfiguration of Point-to-MultiPoint Interfaces
Point-to-MultiPoint interfaces in ATM networks only make sense if no VCs can be set up dynamically because an SVC-capable ATM network normally presents a NBMA cloud to OSPF. This is for example the case if OSPF executes over a network composed of a partial PVC or SPVC mesh or predetermined SVC meshes. Such a network could be modeled using the Point-to-MultiPoint OSPF interface and the neighbor detection could be provided by Proxy-PAR or other means. In the Proxy-PAR case the router queries for all OSPF routers on the same network in the same VPN but it installs in the interface configuration only routers that are already reachable through existing PVCs. The underlying assumption is that a router knows the remote ATM address of a PVC and can compare it with appropriate Proxy-PAR registrations. If the remote ATM address of the PVC is unknown, it can be discovered by such mechanisms as Inverse ARP .
Proxy-PAR provides a true OSPF neighbor detection mechanism, whereas a mechanism like Inverse ARP only returns addresses of directly reachable routers (which are not necessarily running OSPF), in the Point-to-Multi-Point environment.
2.2.3 Autoconfiguration of Numbered Point-to-Point Interfaces
OSPF point-to-point links do not necessarily have an IP address assigned and even if they do, the mask is undefined. As a precondition to successfully register a service with Proxy-PAR, an IP address and a mask are required. Therefore, if a router desires to use Proxy-PAR to advertise the local end of a point-to-point link to the router with which it intends to form an adjacency, an IP address has to be provided as well as a netmask set or a default of 255.255.255.252 (this gives as the default case a subnet with two routers on it) assumed. To allow the discovery of the remote end of the interface, IP address of the remote side has to be provided and a netmask set or a default of 255.255.255.252 assumed. Obviously the discovery can only be successful when both sides of the interface are configured with the same network mask and are within the same IP network. The situation where more than two possible neighbors are discovered through queries and the interface type is set to point- to-point presents a configuration error.
Sending multicast Hello packets on the point-to-point links allows OSPF neighbors to be discovered automatically. On the other hand, using Proxy-PAR instead avoids sending Hello messages to routers that are not necessarily running OSPF.
2.2.4 Autoconfiguration of Unnumbered Point-to-Point Interfaces
For reasons given in , the use of unnumbered point-to-point interfaces with Proxy-PAR is not a very attractive alternative because the lack of an IP address prevents efficient registration and retrieval of configuration information. Relying on the numbering method based on MIB entries generates conflicts with the dynamic nature of creation of such entries and is beyond the scope of this work.
2.3 Registration of OSPF interfaces with Proxy-PAR
To allow other routers to discover an OSPF interface automatically, the IP address, mask, Area ID, interface type and router priority information given must be registered with the Proxy-PAR server at an appropriate scope. A change in any of these parameters has to force a reregistration with Proxy-PAR.
It should be emphasized here that because the registration information can be used by other routers to resolve IP addresses against NSAPs as explained in section 2.4, the entire IP address of the router must be registered. It is not sufficient to indicate the subnet up to the mask length; all address bits must be provided.
2.3.1 Registration of Non-Broadcast Multiple-Access Interfaces
For an NBMA interface the appropriate parameters are available and can be registered through Proxy-PAR without further complications.
2.3.2 Registration of Point-to-Multipoint Interfaces
In the case of a Point-to-MultiPoint interface the router registers its information in the same fashion as in the NBMA case, except that the interface type is modified accordingly.
2.3.3 Registration of Numbered Point-to-Point Interfaces
In the case of point-to-point numbered interfaces the address mask is not specified in the OSPF configuration. If the router has to use Proxy-PAR to advertise its capability, a mask must be defined or a default value of 255.255.255.252 used.
2.3.4 Registration of Unnumbered Point-to-Point Interfaces
Owing to the lack of a configured IP address and difficulties generated by this fact as described earlier, registration of unnumbered point-to-point interfaces is not covered in this document.
2.4 IP address to NSAP Resolution Using Proxy-PAR
As a byproduct of Proxy-PAR presence, an OSPF implementation could use the information in registrations for the resolution of IP addresses to ATM NSAPs on a subnet without having to use static data or mechanisms such as ATMARP . This again should allow a drastic simplification of the number of mechanisms involved in operating OSPF over ATM to provide an IP overlay.
From a system perspective, the OSPF component, the Proxy-PAR client, the IP to NSAP address resolution table, and the ATM circuit manager can be depicted as in Figure 1. Figure 1 shows an example of component interactions triggered by a Proxy-PAR query from the Proxy-PAR client.
2.5 Connection Setup Mechanisms
This section describes the OSPF behavior in an ATM network under various assumptions in terms of signaling capabilities and preset connectivity.
2.5.1 OSPF in PVC Environments
In environments where only partial PVCs (or SPVCs) meshes are available and modeled as Point-to-MultiPoint interfaces, the routers see reachable routers through autodiscovery provided by Proxy-PAR. This leads to expected OSPF behavior. In cases where a full mesh of PVCs is present, such a network should preferably be modeled as NBMA. Note that in such a case, PVCs failures will translate into not-so- obvious routing failures.
__________ _________ | | | | | OSPF |<-------------------|Proxy-PAR|<---(Proxy-PAR query) |__________| notify | client | ^ neighbor changes |_________| | | send and | | maintain Proxy-PAR receive | | entries in table OSPF msg | | | | | | ____V____ ____V_____ | ATM | | | | circuit |-------------------->|IP to NSAP| | manager | check | table | |_________| IP to NSAP bindings |__________|
Figure 1: System perspective of typical components interactions.
2.5.2 OSPF in SVC Environments
+ + + | +---+ | | +--+ |---|RTA|---| +-------+ | +--+ |H1|---| +---+ | | ATM | |---|H2| +--+ | | +---+ | Cloud | +---+ | +--+ |LAN Y |---|RTB|-------------|RTC|---| + | +---+ | PPAR | +---+ | + +-------+ +
Simple topology with Router B and Router C operating across NBMA ATM interfaces with Proxy-PAR.
In SVC-capable environments the routers can initiate VCs after having discovered the appropriate neighbors, preferably driven by the need to send data such as Hello packets. This can lead to race conditions where both sides can open a VC simultaneously. It is generally desirable to avoid wasting this valuable resource: if the router with lower IP address (i.e., the IP address of the OSPF interface registered with Proxy-PAR) detects that the VC initiated by the other side is bidirectional, it is free to close its own VC and use the detected one. Note that this either requires the OSPF implementation to be aware of the VCs used to send and receive Hello messages, or the component responsible of managing VCs to be aware of the usage of particular VCs.
Observe that this behavior operates correctly in case OSPF over Demand Circuits extensions are used  over SVC capable interfaces.
Most of the time, it is possible to avoid the setup of redundant VCs by delaying the sending of the first OSPF Hello from the router with the lower IP address by an amout of time greater than the interval between the queries from the Proxy-PAR client to the server. Chances are that the router with the higher IP address opens the VC (or use an already existing VC) and sends the OSPF Hello first if its interval between queries is shorter than the Hello delay of the router with the lower IP address. As this interval can vary depending on particular needs and implementations, the race conditions described above can still be expected to happen, albeit presumably less often.
The existence of VCs used for OSPF exchanges is orthogonal to the number and type of VCs the router chooses to use within the logical interface to forward data to other routers. OSPF implementations are free to use any of these VCs (in case they are aware of their existence) to send packets if their end points are adequate and must accept Hello packets arriving on any of the VCs belonging to the logical interface even if OSPF operating on such an interface is not aware of their existence. An OSPF implementation may ignore connections being initiated by another router that has not been discovered by Proxy-PAR. In any case, the OSPF implementation will ignore a neighbor whose Proxy-PAR registration indicates that it is not adjacent.
As an example consider the topology in Figure 2 where router RTB and RTC are connected to a common ATM cloud offering Proxy-PAR services. Assuming that RTB's OSPF implementation is aware of SVCs initiated on the interface and that RTC only makes minimal use of Proxy-PAR information, the following sequence could develop, illustrating some of the cases described above:
- RTC and RTB register with ATM cloud as Proxy-PAR capable and discover each other as adjacent OSPF routers.
- RTB sends a Hello, which forces it to establish a SVC connection to RTC.
- RTC sends a Hello to RTB, but disregards the already existing VC and establishes a new VC to RTB to deliver the packet.
- RTB sees a new bidirectional VC and, assuming here that RTC's IP address is higher, closes the VC originated in step 2.
- Host H1 sends data to H2 and RTB establishes a new data SVC between itself and RTC.
- RTB sends a Hello to RTC and decides to do so using the newly establish data SVC. RTC must accept the Hello despite the minimal implementation.
Comments and contributions from several sources, especially Rob Coltun, Doug Dykeman, John Moy and Alex Zinin are included in this work.
4 Security Considerations
Several aspects are to be considered in the context of the security of operating OSPF over ATM and/or Proxy-PAR. The security of registered information handed to the ATM cloud must be guaranteed by the underlying PNNI protocol. The registration itself through Proxy- PAR is not secured, and are thus appropriate mechanisms for further study. However, even if the security at the ATM layer is not guaranteed, OSPF security mechanisms can be used to verify that detected neighbors are authorized to interact with the entity discovering them.
 ATM Forum, "PNNI Augmented Routing (PAR) Version 1.0." ATM Forum af-ra-0104.000, January 1999.  Droz, P. and T. Przygienda, "Proxy-PAR", RFC 2843, May 2000.  ATM-Forum, "Private Network-Network Interface Specification Version 1.0." ATM Forum af-pnni-0055.000, March 1996.  ATM-Forum, "Interim Local Management Interface, (ILMI) Specification 4.0." ATM Forum af-ilmi-0065.000, September 1996.  Laubach, J., "Classical IP and ARP over ATM", RFC 2225, April 1998.  ATM-Forum, "LAN Emulation over ATM 1.0." ATM Forum af-lane- 0021.000, January 1995.  Armitage, G., "Support for Multicast over UNI 3.0/3.1 based ATM Networks", RFC 2022, November 1996.  Coltun, R., "The OSPF Opaque LSA Option", RFC 2328, July 1998.  Davison, M., "ILMI-Based Server Discovery for ATMARP", RFC 2601, June 1999.  Davison, M., "ILMI-Based Server Discovery for MARS", RFC 2602, June 1999.  Davison, M., "ILMI-Based Server Discovery for NHRP", RFC 2603, June 1999.
 Moy, J., "OSPF Version 2", RFC 2328, April 1998.
 Moy, J., "Extending OSPF to Support Demand Circuits", RFC 1793, April 1995.  deSouza, O. and M. Rodrigues, "Guidelines for Running OSPF Over Frame Relay Networks", RFC 1586, March 1994.  Bradley, A. and C. Brown, "Inverse Address Resolution Protocol", RFC 2390, September 1999.
Siara Systems Incorporated
1195 Borregas Avenue
Sunnyvale, CA 94089
Zurich Research Laboratory
Zurich Research Laboratory
Full Copyright Statement
Copyright © The Internet Society (2000). All Rights Reserved.
This document and translations of it may be copied and furnished to others, and derivative works that comment on or otherwise explain it or assist in its implementation may be prepared, copied, published and distributed, in whole or in part, without restriction of any kind, provided that the above copyright notice and this paragraph are included on all such copies and derivative works. However, this document itself may not be modified in any way, such as by removing the copyright notice or references to the Internet Society or other Internet organizations, except as needed for the purpose of developing Internet standards in which case the procedures for copyrights defined in the Internet Standards process must be followed, or as required to translate it into languages other than English.
The limited permissions granted above are perpetual and will not be revoked by the Internet Society or its successors or assigns.
This document and the information contained herein is provided on an "AS IS" basis and THE INTERNET SOCIETY AND THE INTERNET ENGINEERING TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
Funding for the RFC Editor function is currently provided by the Internet Society.