Internet Engineering Task Force (IETF)
Request for Comments: 6760
Category: Informational
ISSN: 2070-1721
S. Cheshire
M. Krochmal
Apple Inc.
February 2013

Requirements for a Protocol to Replace

the AppleTalk Name Binding Protocol (NBP)

Abstract

One of the goals of the authors of Multicast DNS (mDNS) and DNS-Based Service Discovery (DNS-SD) was to retire AppleTalk and the AppleTalk Name Binding Protocol (NBP) and to replace them with an IP-based solution. This document presents a brief overview of the capabilities of AppleTalk NBP and outlines the properties required of an IP-based replacement.

Status of This Memo

This document is not an Internet Standards Track specification; it is published for informational purposes.

This document is a product of the Internet Engineering Task Force (IETF). It represents the consensus of the IETF community. It has received public review and has been approved for publication by the Internet Engineering Steering Group (IESG). Not all documents approved by the IESG are a candidate for any level of Internet Standard; see Section 2 of RFC 5741.

Information about the current status of this document, any errata, and how to provide feedback on it may be obtained at http://www.rfc-editor.org/info/rfc6760.

Copyright Notice

Copyright © 2013 IETF Trust and the persons identified as the document authors. All rights reserved.

This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.

Table of Contents

   1. Introduction ....................................................3
   2. Zero Configuration Networking ...................................4
   3. Requirements ....................................................4
      3.1. Name-to-Address Mapping ....................................5
      3.2. Name Services, Not Hardware ................................5
      3.3. Address Services, Not Hardware -- or -- Escape the
           Tyranny of Well-Known Ports ................................6
      3.4. Typed Name Space ...........................................8
      3.5. User-Friendly Names ........................................9
      3.6. Zeroconf Operation .........................................9
      3.7. Name Space Management -- or -- Name Conflict Detection ....10
      3.8. Late Binding ..............................................11
      3.9. Simplicity ................................................11
      3.10. Network Browsing .........................................11
      3.11. Browsing and Registration Guidance .......................12
      3.12. Power Management Support .................................12
      3.13. Protocol Agnostic ........................................13
      3.14. Distributed Cache Coherency Protocol .....................13
      3.15. Immediate and Ongoing Information Presentation ...........13
   4. Existing Protocols .............................................14
   5. IPv6 Considerations ............................................14
   6. Security Considerations ........................................14
   7. Informative References .........................................15

1. Introduction

An important goal of the participants working on Zeroconf, Multicast DNS, and DNS-Based Service Discovery was to provide a viable IP-based replacement for AppleTalk and the AppleTalk Name Binding Protocol (NBP).

There are many who are experts in the Domain Name System (DNS) who know nothing about the AppleTalk Name Binding Protocol (NBP). Without some background on how AppleTalk and NBP worked, it may be difficult to understand the reasoning and motivations that led to the design decisions in Multicast DNS and DNS-Based Service Discovery.

This document seeks to remedy this problem by clearly stating the requirements for an IP-based replacement for AppleTalk and NBP. Replacing NBP was not the sole goal of Multicast DNS; therefore, these requirements are not the sole design considerations. However, replacing NBP was a major motivation behind the work in Multicast DNS.

In most cases, the requirements presented in this document are simply a restatement of what AppleTalk NBP currently does. However, this document is not restricted to describing only what NBP currently does. Achieving at least equivalent functionality to NBP is a necessary but not sufficient condition for a viable replacement. In some cases, the requirements for a viable IP-based replacement go beyond NBP. For example, AppleTalk NBP uses Apple Extended ASCII for its character set. It is clear that an IP-based replacement being designed today should use Unicode, in the form of UTF-8 [RFC3629]. AppleTalk NBP has a reputation, partially deserved, for being too 'chatty' on the network. An IP-based replacement should not have this same failing. The intent is to learn from NBP and build a superset of its functionality, not to replicate it precisely with all the same flaws.

The protocols specified in "Multicast DNS" [RFC6762] and "DNS-Based Service Discovery" [RFC6763], taken together, describe a solution that meets these requirements. This document is written, in part, in response to requests for more background information explaining the rationale behind the design of those protocols.

2. Zero Configuration Networking

Historically, TCP/IP networking required configuration, either in the form of manual configuration by a human operator or in the form of automated configuration provided by a DHCP server [RFC2131].

One of the characteristics of AppleTalk was that it could operate without any dependency on manual configuration or a network service to provide automated configuration. An AppleTalk network could be as small as just two laptop computers connected via an Ethernet cable (or wirelessly).

IP now has self-assigned link-local addresses [RFC3927] [RFC4862], which enable IP-based networking in the absence of external configuration. What remains is the need for Zero Configuration name- to-address translation and Zero Configuration service discovery, both capabilities that AppleTalk NBP offered.

It is not necessarily the case that Zero Configuration Networking protocols will always be used in all three areas (addressing, naming, and service discovery) simultaneously on any given network. For example, even on networks with a DHCP server to provide address configuration, users may still use Zero Configuration protocols for name-to-address translation and service discovery. Indeed, on a single network, users may use conventional Unicast DNS for looking up the addresses of Internet web sites while at the same time using Multicast DNS for looking up the addresses of peers on the local link. Therefore, Zero Configuration Networking protocols must coexist peacefully with conventional configured IP networking when used together on the same network.

Networks change state over time. Hosts and services may come and go. Connectivity, addresses, and names change. In a manually configured network, a human operator can remedy errors when they arise. In a Zero Configuration Network, no such human operator is available to diagnose and troubleshoot problems, so Zero Configuration protocols need to be self-correcting, automatically accommodating changing network conditions, continually converging to correctness in the absence of human intervention to manually rectify errors.

3. Requirements

This section lists the 15 requirements for an IP-based replacement for AppleTalk NBP.

3.1. Name-to-Address Mapping

NBP's primary function is translating names to addresses.

NBP stands for Name Binding Protocol, not Network Browsing Protocol. Many people know NBP only as "that thing that used to let you browse the network in the old Macintosh Chooser". While browsing is an important facility of NBP, it is secondary to NBP's primary function of translating names to addresses.

Every time a user prints using AppleTalk, the printing software takes the name of the currently selected printer, looks up the current AppleTalk address associated with that named service, and establishes a connection to that service on the network. The user may invoke NBP's browsing capability once, when first selecting the desired printer in the Chooser, but after that, every time something is printed, it is a simple efficient name-to-address lookup that is being performed, not a full-fledged browsing operation.

Any NBP replacement needs to support, as its primary function, an efficient name-to-address lookup operation.

3.2. Name Services, Not Hardware

The primary named entities in NBP are services, not "hosts", "machines", "devices", or pieces of hardware of any kind. This concept is more subtle than it may seem at first, so it bears some discussion.

The AppleTalk NBP philosophy is that naming a piece of hardware on the network is of little use if you can't communicate with that piece of hardware. To communicate with a piece of hardware, there needs to be a piece of software running on that hardware that sends and receives network packets conforming to some specific protocol. This means that whenever you communicate with a machine, you are really communicating with some piece of software on that machine. Even if you just 'ping' a machine to see if it is responding, it is not really the machine that you are 'pinging', it is the software on that machine that generates ICMP Echo Responses [RFC792].

Consequently, this means that the only things worth naming are the software entities with which you can communicate. A user who wants to use a print server or a file server needn't care about what hardware implements those services. There may be a single machine hosting both services, or there may be two separate machines. The end user doesn't need to care.

The one exception to this is network managers, who may want to name physical hardware for the purpose of tracking physical inventory. However, even this can be recast into a service-oriented view of the world by saying that what you're naming is not the hardware, but the ICMP Echo Responder that runs (or is assumed to be running) on every piece of IP hardware.

3.3. Address Services, Not Hardware -- or -- Escape the Tyranny of

Well-Known Ports

The reader may argue that DNS already supports the philosophy of naming services instead of hosts. When we see names like "www.example.com.", "pop.example.com.", "smtp.example.com.", "news.example.com.", and "time.example.com.", we do not assume that those names necessarily refer to different hosts. They are clearly intended to be logical service names and could, in fact, all resolve to the same IP address.

The shortcoming here is that although the names are clearly logical service names, the result today of doing a conventional ("A" or "AAAA") DNS lookup for those names gives you only the IP address of the hardware where the service is located. To communicate with the desired service, you also need to know the port number at which the service can be reached, not just the IP address.

This means that the port number has to be communicated out-of-band, in some other way. One way is for the port number to be a specific well-known constant for any given protocol. This makes it hard to run more than one instance of a service on a single piece of hardware. Another way is for the user to explicitly type in the port number, for example, "www.example.com.:8080" instead of "www.example.com.", but needing to know and type in a port number is as ugly and fragile as needing to know and type in an IP address.

Another aspect of the difficulty of running more than one instance of a service on a single piece of hardware is that it forces application programmers to write their own demultiplexing capability. AppleTalk did not suffer this limitation. If an AppleTalk print server offered three print queues, each print queue ran as its own independent service, listening on its own port number (called a socket number in AppleTalk terminology), each advertised as a separate, independently named NBP entity. When a client looks up the address of that named NBP entity, the reply encodes not only on which net and subnet the service resides, and on which host on that subnet (like an IP address does), but also on which socket number (port number) within that host. In contrast, if an lpr print server offers three print queues, all three print queues are typically reached through the same well- known port number (515), and then the lpr protocol has to use its own demultiplexing capability (the print queue name) in order to determine which print queue is sought. This makes it especially difficult to run two different pieces of print queue software from different vendors on the same machine, because they cannot both listen on the same well-known port.

A similar trick is used in HTTP 1.1, where the "Host" header line is used to allow multiple logical HTTP services to run at the same IP address. Again, this works for a single-vendor solution, but if a user wishes to run multiple web servers (for example, an image server, a database program, an HTTP email access gateway, and a conventional HTTP server) on a single machine, they can't all listen on TCP port 80, the traditional HTTP port.

Yet another problem of well-known ports is that port numbers are a finite resource. Originally, port numbers 0-255 were reserved for well-known services, and the remaining 99.6% of the port space was free for dynamic allocation [RFC1122]. Since then, the range of "Registered Ports" has crept upwards until today, ports 0-49151 are reserved, and only 25% of the space remains available for dynamic allocation. Even though 65535 may seem like a lot of available port numbers, with the pace of software development today, if every new protocol gets its own private port number, we will eventually run out. To avoid having to do application-level demultiplexing, protocols like the X Window System wisely use a range of port numbers, and let TCP do the demultiplexing for them. The X Window System uses 64 ports, in the range 6000-6063. If every new protocol were to get its own chunk of 64 ports, we would run out even faster.

Any NBP replacement needs to provide, not just the network number, subnet number, and host number within that subnet (i.e., the IP address) but also the port number within that host where the service is located. Furthermore, since many existing IP services such as lpr *do* already use additional application-layer demultiplexing information such as a print queue name, an NBP replacement needs to support this too by including this information as part of the complete package of addressing information provided to the client to enable it to use the service. The NBP replacement needs to name individual print queues as first-class entities in their own right. It is not sufficient merely to name a print server, within which separate print queues can then be found by some other mechanism.

One possible answer here is that an IP-based NBP replacement could use a solution derived from DNS SRV records instead of "A" records, since SRV records *do* provide a port number. However, this alone is not a complete solution, because SRV records cannot tell you an lpr print queue name.

3.4. Typed Name Space

AppleTalk NBP names are structured names, generally written as:

      Name : Type @ Zone

Name:

The Name is the user-visible name of the service.

Type: The Type is an opaque identifier that identifies the service protocol and semantics. The user may think of the Type as identifying the end-user function that the device performs (e.g., "printing"), and for the typical end-user, this may be an adequate mental model, but strictly speaking, from a protocol-design perspective, the Type identifies the semantic application protocol the service speaks: no more, no less. For convenience, the opaque Type identifier is generally constructed using descriptive ASCII text, but this text has no meaning to the protocol, and care should be taken in inferring too much meaning from it. For example, the NBP Service Type "LaserWriter" means "any service that speaks PS/PAP/ATP/DDP (PostScript over AppleTalk Printer Access Protocol over AppleTalk Transaction Protocol over AppleTalk Datagram Delivery Protocol)". It does not necessarily mean an Apple-branded "LaserWriter" printer; nor does the service even have to be a printer. A device that archives documents to digital media could advertise itself as a "LaserWriter", meaning that it speaks PostScript over PAP, not necessarily that it prints that document on paper when it gets it. The end-user never directly sees the Service Type. It is implicit in the user's action; for example, when printing, the printing software knows what protocol(s) it speaks and consequently what Service Type(s) it should be looking for -- the user doesn't have to tell it.

Zone: The Zone is an organizational or geographical grouping of named services. AppleTalk Zones were typically given names like "Engineering", "Sales", or "Building 1, 3rd floor, North". The equivalent concept in DNS could be a subdomain such as "Engineering.example.com.", "Sales.example.com.", or "Building 1, 3rd floor, North.example.com."

Each {Type,Zone} pair defines a name space in which service names can be registered. It is not a name conflict to have a printer called "Sales" and a file server called "Sales", because one is "Sales:LaserWriter@Zone" and the other is "Sales:AFPServer@Zone".

Any NBP replacement needs to provide a mechanism that allows names to be grouped into organizational or geographical "zones", and within each "zone", to provide an independent name space for each service type.

3.5. User-Friendly Names

When repeatedly typing in names on command-line systems, it is helpful to have names that are short, all lowercase, without spaces or hard-to-type characters.

Since Service Names are intended to be selected from a list, not repeatedly typed in on a keyboard, there is no reason for them to be restricted so. Users should be able to give their printers names like "Sales", "Marketing", and "3rd Floor Copy Room", not just "printer1.example.com.". Of course, a user is free to name a particular service using only lowercase letters and no spaces if they wish, but they should not be forced to do that.

Any NBP replacement needs to support a full range of rich text characters, including uppercase, lowercase, spaces, accented characters, and so on. The correct solution is likely to be UTF-8 Unicode [RFC3629].

Note that this requirement for user-friendly rich-text names applies equally to the zones (domains) in which services are registered and discovered.

Note that although the characters ':' and '@' are used when writing AppleTalk NBP names, they are simply a notational convenience in written text. In the on-the-wire protocol and in the software data structures, NBP Name, Type, and Zone strings are all allowed to contain almost any character, including ':' and '@'. The naming scheme provided by an NBP replacement must allow the use of any desired characters in service names, including dots ('.'), spaces, percent signs, etc.

3.6. Zeroconf Operation

AppleTalk NBP is self-configuring. On a network of just two hosts, they communicate peer-to-peer using multicast. On a large managed network, AppleTalk routers automatically perform an aggregation function, allowing name lookups to be performed via unicast to a service running on the router, instead of by flooding the entire network with multicast packets to every host.

Any NBP replacement needs to be able to operate in the absence of external network infrastructure. However, this should not be the only mode of operation. In larger managed networks, it should also be possible to take advantage of appropriate external network infrastructure when present, to perform queries via unicast instead of multicast.

3.7. Name Space Management -- or -- Name Conflict Detection

Because an NBP replacement needs to operate in a Zeroconf environment, it cannot be assumed that a central network administrator is managing the network. Unlike managed networks where normal administrative controls may apply, in the Zeroconf case an NBP replacement must make it easy for users to name their devices as they wish, without the inconvenience or expense of having to seek permission or pay some organization like a domain name registry for the privilege. However, this ease of naming, and freedom to choose any desired name, may lead to name conflicts. Two users may independently decide to run a personal file server on their laptop computers, and (unimaginatively) name it "My Computer". When these two users later attend the next IETF meeting and find themselves part of the same wireless network, there may be problems.

Similarly, every Brother network printer may ship from the factory with its Service Name set to "Brother Printer". On a typical small home network where there is only one printer, this is not a problem; however, it could become a problem if two or more such printers are connected to the same network.

Any NBP replacement needs to detect such conflicts, and handle them appropriately. In the case of a laptop computer, which has a keyboard, screen, and a human user, the software should display a message telling the user that they need to select a new name.

In the case of printers, which typically have no keyboard or screen, the software should automatically select a new unique name, perhaps by appending an integer to the end of the existing name, e.g., "Brother Printer 2". Note that, although this programmatically derived name should be recorded persistently for use next time the device is powered on, the user is not forced to use that name as the long-term name for the service/device. In a network with more than one printer, the typical user will assign human-meaningful names to those printers, such as "Upstairs Printer" and "Downstairs Printer", but the ability to rename the printer using some configuration tool (e.g., a web browser) depends on the ability to find the printer and connect to it in the first place. Hence, the programmatically derived unique name serves a vital bootstrapping role, even if its use in that role is temporary.

Because of the potentially transient nature of connectivity on small portable devices that are becoming more and more common (especially when used with wireless networks), this name conflict detection needs to be an ongoing process. It is not sufficient simply to verify uniqueness of names for a few seconds during the boot process and then assume that the names will remain unique indefinitely.

If the Zeroconf naming mechanism is integrated with the existing global DNS naming mechanism, then it would be beneficial for a sub- tree of that global namespace to be designated as having only local significance, for use without charge by cooperating peers, much as portions of the IPv4 address space are already designated as local- significance-only, available for organizations to use locally without charge as they wish [RFC1918].

3.8. Late Binding

When the user selects their default printer, the software should store only the name, not the IP address and port number. Then, every time the user prints, the software should look up the name to find the current IP address and port number for that service. This allows a named logical service to be moved from one piece of hardware to another without disrupting the user's ability to print to that named print service.

On a network using DHCP [RFC2131] or self-assigned link-local addresses [RFC3927] [RFC4862], a device's IP address may change from day to day. Deferring binding of name to address until actual use allows the client to get the correct IP address at the time the service is used.

Similarly, with a service using a dynamic port number instead of a fixed well-known port, the service may not get the same port number every time it is started or restarted. By deferring binding of name to port number until actual use, the client gets the correct port number at the time the service is used.

3.9. Simplicity

Any NBP replacement needs to be simple enough that vendors of even a low-cost network ink-jet printer can afford to implement it in the device's limited firmware.

3.10. Network Browsing

AppleTalk NBP offers certain limited wild-card functionality. For example, the service name "=" means "any name". This allows a client to perform an NBP lookup such as "=:LaserWriter@My Zone" and receive back in response a list of all the PS/PAP (AppleTalk Printer Access Protocol) printers in the Zone called "My Zone".

Any NBP replacement needs to allow a piece of software, such as a printing client or a file server client, to enumerate all the named instances of services in a specified zone (domain) that speak its protocol(s).

3.11. Browsing and Registration Guidance

AppleTalk NBP provides certain meta-information to the client.

On a network with multiple AppleTalk Zones, the AppleTalk network infrastructure informs the client of the list of Zones that are available for browsing. It also informs the client of the default Zone, which defines the client's logical "home" location. This is the Zone that is selected by default when the Macintosh Chooser is opened, and is usually the Zone where the user is most likely to find services like printers that are physically nearby, but the user is still free to browse any Zone in the offered list that they wish.

A Brother printer may be pre-configured at the factory with the Service Name "Brother Printer", but they do not know on which network the printer will eventually be installed, so the printer will have to learn this from the network on arrival. On a network with multiple AppleTalk Zones, the AppleTalk network infrastructure informs the client of a single default Zone within which it may register Service Names. In the case of a device with a human user, the AppleTalk network infrastructure may also inform the client of a list of Zones within which the client may register Service Names, and the user may choose to register Service Names in any one of those Zones instead of in the suggested default Zone.

Any NBP replacement needs to provide the following information to the client:

  • The suggested zone (domain) in which to register Service Names.
  • A list of recommended available zones (domains) in which Service Names may be optionally registered.
  • The suggested default zone (domain) for network browsing.
  • A list of available zones (domains) that may be browsed.

Note that, because the domains used in this context are intended for service browsing in a graphical user interface, they should be permitted to be full user-friendly rich text, just like the rest of a service name.

3.12. Power Management Support

Many modern network devices have the ability to go into a low-power mode, where only a small part of the Ethernet hardware remains powered, and the device can be woken up by sending a specially formatted Ethernet frame that the device's power-management hardware recognizes. A modern service discovery protocol should provide facilities to enable this low-power mode to be used effectively without sacrificing network functionality, such as the ability to discover services on sleeping devices, and wake up a sleeping device when it is needed.

3.13. Protocol Agnostic

Fashions come and go in the computer industry, but a service discovery protocol, being one of the foundation components on which everything else rests, has to be able to outlive these swings of fashion. A useful service discovery protocol should be agnostic to the protocols being used by the higher-layer software it serves. If a service discovery protocol requires all the higher-layer software to be written in a new computer language, or requires all the higher- layer protocols to embrace some trendy new data representation format that is currently in vogue, then that service discovery protocol is likely to have limited utility after the fashion changes and computer industry moves on to its next infatuation.

3.14. Distributed Cache Coherency Protocol

Any modern service discovery protocol must use some kind of caching for efficiency. Any time a distributed cache is maintained, a cache coherency protocol is required to control the effects of stale data. Thus, a useful service discovery protocol needs to include cache coherency mechanisms.

3.15. Immediate and Ongoing Information Presentation

Many current discovery mechanisms display an hourglass or a "Please Wait" message for five or ten seconds, and then present a list of results to the user. At this point, the list of results is static, and does not update in response to changes in the environment. To see current information, the user is forced to click a "Refresh" button repeatedly, waiting another five to ten seconds each time.

Neither limitation is acceptable in a protocol that is to replace NBP. When a user initiates a browsing operation, the user interface should take at most one second to present the list of results. In addition, the list should update in response to changes in the environment as they happen. If the user is waiting for a particular service to become available, they should be able simply to watch until it appears, with no "Refresh" button that they need to keep clicking. A protocol to replace AppleTalk NBP must be able to meet these requirement for timeliness of information discovery, and liveness of information updating, without placing undue burden on the network.

4. Existing Protocols

Ever since this work began with Stuart Cheshire's email to the net- thinkers@thumper.vmeng.com mailing list in July 1997, the question has been asked, "Isn't SLP the IETF replacement for AppleTalk NBP?"

The Service Location Protocol (SLP) [RFC2608] provides extremely rich and flexible facilities in the area of Requirement 10, "Network Browsing". However, SLP provides none of the service naming, automatic name conflict detection, or efficient name-to-address lookup that form the majority of what AppleTalk NBP does.

SLP returns results in the form of URLs. In the absence of DNS, URLs cannot usefully contain DNS names. Discovering a list of service URLs of the form "ipp://169.254.17.202/" is not particularly informative to the user. Discovering a list of service URLs of the form "ipp://epson-stylus-900n.local./" is slightly less opaque (though still not very user-friendly), but to do even this, SLP would have to depend on Multicast DNS or something similar to resolve names to addresses in the absence of a conventional DNS server.

SLP provides fine-grained query capabilities, such as the ability to prune a long list of printers to show only those that have blue paper in the top tray, which could be useful on extremely large networks with very many printers, but are certainly unnecessary for a typical home or small office with only one or two printers.

In summary, SLP alone fails to meet most of the requirements, and provides vastly more mechanism than necessary in the area of Requirement 10.

5. IPv6 Considerations

An IP replacement for the AppleTalk Name Binding Protocol needs to support IPv6 as well as IPv4.

6. Security Considerations

The AppleTalk Name Binding Protocol was developed in an era when little consideration was given to security issues. In today's world, this would no longer be appropriate. Any modern replacement for AppleTalk NBP should have security measures appropriate to the environment in which it will be used. Given that this document is a broad historical overview of how AppleTalk NBP worked, and does not specify any new protocol(s), it is beyond the scope of this document to provide detailed discussion of possible network environments, what protocols would be appropriate in each, and what security measures would be expected of each such protocol.

7. Informative References

   [RFC792]   Postel, J., "Internet Control Message Protocol", STD 5,
              RFC 792, September 1981.
   
   [RFC1122]  Braden, R., Ed., "Requirements for Internet Hosts -
              Communication Layers", STD 3, RFC 1122, October 1989.
   
   [RFC1918]  Rekhter, Y., Moskowitz, B., Karrenberg, D., de Groot, G.,
              and E. Lear, "Address Allocation for Private Internets",
              BCP 5, RFC 1918, February 1996.
   
   [RFC2131]  Droms, R., "Dynamic Host Configuration Protocol", RFC
              2131, March 1997.
   
   [RFC2608]  Guttman, E., Perkins, C., Veizades, J., and M. Day,
              "Service Location Protocol, Version 2", RFC 2608, June
              1999.
   
   [RFC3629]  Yergeau, F., "UTF-8, a transformation format of ISO
              10646", STD 63, RFC 3629, November 2003.
   
   [RFC3927]  Cheshire, S., Aboba, B., and E. Guttman, "Dynamic
              Configuration of IPv4 Link-Local Addresses", RFC 3927, May
              2005.
   
   [RFC4862]  Thomson, S., Narten, T., and T. Jinmei, "IPv6 Stateless
              Address Autoconfiguration", RFC 4862, September 2007.
   
   [RFC6762]  Cheshire, S. and M. Krochmal, "Multicast DNS", RFC 6762,
              February 2013.
   
   [RFC6763]  Cheshire, S. and M. Krochmal, "DNS-Based Service
              Discovery", RFC 6763, February 2013.

Authors' Addresses

   Stuart Cheshire
   Apple Inc.
   1 Infinite Loop
   Cupertino, CA  95014
   USA
   
   Phone: +1 408 974 3207
   EMail: cheshire@apple.com
   
   Marc Krochmal
   Apple Inc.
   1 Infinite Loop
   Cupertino, CA  95014
   USA
   
   Phone: +1 408 974 4368
   EMail: marc@apple.com