[RFC1035] Mockapetris, P., "Domain names - implementation and
specification", STD 13, RFC 1035, November 1987.
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119, March 1997.
[RFC2782] Gulbrandsen, A., Vixie, P., and L. Esibov, "A DNS RR for
specifying the location of services (DNS SRV)", RFC 2782,
[RFC3492] Costello, A., "Punycode: A Bootstring encoding of Unicode
for Internationalized Domain Names in Applications
(IDNA)", RFC 3492, March 2003.
[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,
[RFC4862] Thomson, S., Narten, T., and T. Jinmei, "IPv6 Stateless
Address Autoconfiguration", RFC 4862, September 2007.
[RFC5198] Klensin, J. and M. Padlipsky, "Unicode Format for Network
Interchange", RFC 5198, March 2008.
[RFC5890] Klensin, J., "Internationalized Domain Names for
Applications (IDNA): Definitions and Document Framework",
RFC 5890, August 2010.
[RFC6335] Cotton, M., Eggert, L., Touch, J., Westerlund, M., and S.
Cheshire, "Internet Assigned Numbers Authority (IANA)
Procedures for the Management of the Service Name and
Transport Protocol Port Number Registry", BCP 165, RFC
6335, August 2011.
18.2. Informative References
[AFP] Mac OS X Developer Library, "Apple Filing Protocol
Programming Guide", <http://developer.apple.com/
[BJ] Apple Bonjour Open Source Software,
Appendix A. Rationale for Using DNS as a Basis for Service Discovery
Over the years, there have been many proposed ways to do network
service discovery with IP, but none achieved ubiquity in the
marketplace. Certainly none has achieved anything close to the
ubiquity of today's deployment of DNS servers, clients, and other
The advantage of using DNS as the basis for service discovery is that
it makes use of those existing servers, clients, protocols,
infrastructure, and expertise. Existing network analyzer tools
already know how to decode and display DNS packets for network
For ad hoc networks such as Zeroconf environments, peer-to-peer
multicast protocols are appropriate. Using DNS-SD running over
Multicast DNS [RFC6762] provides zero-configuration ad hoc service
discovery, while maintaining the DNS-SD semantics and record types
In larger networks, a high volume of enterprise-wide IP multicast
traffic may not be desirable, so any credible service discovery
protocol intended for larger networks has to provide some facility to
aggregate registrations and lookups at a central server (or servers)
instead of working exclusively using multicast. This requires some
service discovery aggregation server software to be written,
debugged, deployed, and maintained. This also requires some service
discovery registration protocol to be implemented and deployed for
clients to register with the central aggregation server. Virtually
every company with an IP network already runs a DNS server, and DNS
already has a dynamic registration protocol [RFC2136] [RFC3007].
Given that virtually every company already has to operate and
maintain a DNS server anyway, it makes sense to take advantage of
this expertise instead of also having to learn, operate, and maintain
a different service registration server. It should be stressed again
that using the same software and protocols doesn't necessarily mean
using the same physical piece of hardware. The DNS-SD service
discovery functions do not have to be provided by the same piece of
hardware that is currently providing the company's DNS name service.
The "_tcp.<Domain>" and "_udp.<Domain>" subdomains may be delegated
to a different piece of hardware. However, even when the DNS-SD
service is being provided by a different piece of hardware, it is
still the same familiar DNS server software, with the same
configuration file syntax, the same log file format, and so forth.
Service discovery needs to be able to provide appropriate security.
DNS already has existing mechanisms for security [RFC4033].
Service discovery requires a central aggregation server.
DNS already has one: a DNS server.
Service discovery requires a service registration protocol.
DNS already has one: DNS Dynamic Update.
Service discovery requires a query protocol.
DNS already has one: DNS queries.
Service discovery requires security mechanisms.
DNS already has security mechanisms: DNSSEC.
Service discovery requires a multicast mode for ad hoc networks.
Using DNS-SD in conjunction with Multicast DNS provides this,
using peer-to-peer multicast instead of a DNS server.
It makes more sense to use the existing software that every network
needs already, instead of deploying an entire parallel system just
for service discovery.
Appendix B. Ordering of Service Instance Name Components
There have been questions about why services are named using DNS
Service Instance Names of the form:
Service Instance Name = <Instance> . <Service> . <Domain>
Service Instance Name = <Service> . <Instance> . <Domain>
There are three reasons why it is beneficial to name service
instances with the parent domain as the most-significant (rightmost)
part of the name, then the abstract service type as the next-most
significant, and then the specific instance name as the least-
significant (leftmost) part of the name. These reasons are discussed
below in Sections B.1, B.2, and B.3.
B.1. Semantic Structure
The facility being provided by browsing ("Service Instance
Enumeration") is effectively enumerating the leaves of a tree
structure. A given domain offers zero or more services. For each of
those service types, there may be zero or more instances of that
The user knows what type of service they are seeking. (If they are
running an FTP client, they are looking for FTP servers. If they
have a document to print, they are looking for entities that speak
some known printing protocol.) The user knows in which
organizational or geographical domain they wish to search. (The user
does not want a single flat list of every single printer on the
planet, even if such a thing were possible.) What the user does not
know in advance is whether the service they seek is offered in the
given domain, or if so, the number of instances that are offered and
the names of those instances.
Hence, having the instance names be the leaves of the tree is
consistent with this semantic model.
Having the service types be the terminal leaves of the tree would
imply that the user knows the domain name and the name of the service
instance, but doesn't have any idea what the service does. We would
argue that this is a less useful model.
B.2. Network Efficiency
When a DNS response contains multiple answers, name compression works
more effectively if all the names contain a common suffix. If many
answers in the packet have the same <Service> and <Domain>, then each
occurrence of a Service Instance Name can be expressed using only the
<Instance> part followed by a two-byte compression pointer
referencing a previous appearance of "<Service>.<Domain>". This
efficiency would not be possible if the <Service> component appeared
first in each name.
B.3. Operational Flexibility
This name structure allows subdomains to be delegated along logical
service boundaries. For example, the network administrator at
Example Co. could choose to delegate the "_tcp.example.com."
subdomain to a different machine, so that the machine handling
service discovery doesn't have to be the machine that handles other
day-to-day DNS operations. (It *can* be the same machine if the
administrator so chooses, but the administrator is free to make that
choice.) Furthermore, if the network administrator wishes to
delegate all information related to IPP printers to a machine
dedicated to that specific task, this is easily done by delegating
the "_ipp._tcp.example.com." subdomain to the desired machine. It is
also convenient to set security policies on a per-zone/per-subdomain
basis. For example, the administrator may choose to enable DNS
Dynamic Update [RFC2136] [RFC3007] for printers registering in the
"_ipp._tcp.example.com." subdomain, but not for other
zones/subdomains. This easy flexibility would not exist if the
<Service> component appeared first in each name.
Appendix C. What You See Is What You Get
Some service discovery protocols decouple the true service identifier
from the name presented to the user. The true service identifier
used by the protocol is an opaque unique identifier, often
represented using a long string of hexadecimal digits, which should
never be seen by the typical user. The name presented to the user is
merely one of the decorative ephemeral attributes attached to this
The problem with this approach is that it decouples user perception
from network reality:
* What happens if there are two service instances, with different
unique ids, but they have inadvertently been given the same user-
visible name? If two instances appear in an on-screen list with
the same name, how does the user know which is which?
* Suppose a printer breaks down, and the user replaces it with
another printer of the same make and model, and configures the new
printer with the exact same name as the one being replaced:
"Stuart's Printer". Now, when the user tries to print, the on-
screen print dialog tells them that their selected default printer
is "Stuart's Printer". When they browse the network to see what
is there, they see a printer called "Stuart's Printer", yet when
the user tries to print, they are told that the printer "Stuart's
Printer" can't be found. The hidden internal unique identifier
that the software is trying to find on the network doesn't match
the hidden internal unique identifier of the new printer, even
though its apparent "name" and its logical purpose for being there
are the same. To remedy this, the user typically has to delete
the print queue they have created, and then create a new
(apparently identical) queue for the new printer, so that the new
queue will contain the right hidden internal unique identifier.
Having all this hidden information that the user can't see makes
for a confusing and frustrating user experience, and exposing
long, ugly hexadecimal strings to the user and forcing them to
understand what they mean is even worse.
* Suppose an existing printer is moved to a new department, and
given a new name and a new function. Changing the user-visible
name of that piece of hardware doesn't change its hidden internal
unique identifier. Users who had previously created a print queue
for that printer will still be accessing the same hardware by its
unique identifier, even though the logical service that used to be
offered by that hardware has ceased to exist.
Solving these problems requires the user or administrator to be aware
of the supposedly hidden unique identifier, and to set its value
correctly as hardware is moved around, repurposed, or replaced,
thereby contradicting the notion that it is a hidden identifier that
human users never need to deal with. Requiring the user to
understand this expert behind-the-scenes knowledge of what is
*really* going on is just one more burden placed on the user when
they are trying to diagnose why their computers and network devices
are not working as expected.
These anomalies and counterintuitive behaviors can be eliminated by
maintaining a tight bidirectional one-to-one mapping between what the
user sees on the screen and what is really happening "behind the
curtain". If something is configured incorrectly, then that is
apparent in the familiar day-to-day user interface that everyone
understands, not in some little-known, rarely used "expert"
In summary: in DNS-SD the user-visible name is also the primary
identifier for a service. If the user-visible name is changed, then
conceptually the service being offered is a different logical service
-- even though the hardware offering the service may have stayed the
same. If the user-visible name doesn't change, then conceptually the
service being offered is the same logical service -- even if the
hardware offering the service is new hardware brought in to replace
some old equipment.
There are certainly arguments on both sides of this debate.
Nonetheless, the designers of any service discovery protocol have to
make a choice between having the primary identifiers be hidden, or
having them be visible, and these are the reasons that we chose to
make them visible. We're not claiming that there are no
disadvantages of having primary identifiers be visible. We
considered both alternatives, and we believe that the few
disadvantages of visible identifiers are far outweighed by the many
problems caused by use of hidden identifiers.
Appendix D. Choice of Factory-Default Names
When a DNS-SD service is advertised using Multicast DNS [RFC6762], if
there is already another service of the same type advertising with
the same name then automatic name conflict resolution will occur. As
described in the Multicast DNS specification [RFC6762], upon
detecting a conflict, the service should:
1. Automatically select a new name (typically by appending or
incrementing a digit at the end of the name),
2. Try advertising with the new name, and
3. Upon success, record the new name in persistent storage.
This renaming behavior is very important, because it is key to
providing user-friendly instance names in the out-of-the-box factory-
default configuration. Some product developers apparently have not
realized this, because there are some products today where the
factory-default name is distinctly unfriendly, containing random-
looking strings of characters, such as the device's Ethernet address
in hexadecimal. This is unnecessary and undesirable, because the
point of the user-visible name is that it should be friendly and
meaningful to human users. If the name is not unique on the local
network then the protocol will remedy this as necessary. It is
ironic that many of the devices with this design mistake are network
printers, given that these same printers also simultaneously support
AppleTalk-over-Ethernet, with nice user-friendly default names (and
automatic conflict detection and renaming). Some examples of good
factory-default names are:
HP LaserJet 4600
Ricoh Aficio CL7100
Xerox Phaser 6200DX
To make the case for why adding long, ugly factory-unique serial
numbers to the end of names is neither necessary nor desirable,
consider the cases where the user has (a) only one network printer,
(b) two network printers, and (c) many network printers.
(a) In the case where the user has only one network printer,
a simple name like (to use a vendor-neutral example)
"Printer" is more user-friendly than an ugly name like
"Printer_0001E68C74FB". Appending ugly hexadecimal goop to the
end of the name to make sure the name is unique is irrelevant to
a user who only has one printer anyway.
(b) In the case where the user gets a second network printer, having
the new printer detect that the name "Printer" is already in use
and automatically name itself "Printer (2)" instead, provides a
good user experience. For most users, remembering that the old
printer is "Printer" and the new one is "Printer (2)" is easy
and intuitive. Seeing a printer called "Printer_0001E68C74FB"
and another called "Printer_00306EC3FD1C" is a lot less helpful.
(c) In the case of a network with ten network printers, seeing a
list of ten names all of the form "Printer_xxxxxxxxxxxx" has
effectively taken what was supposed to be a list of user-
friendly rich-text names (supporting mixed case, spaces,
punctuation, non-Roman characters, and other symbols) and turned
it into just about the worst user interface imaginable: a list
of incomprehensible random-looking strings of letters and
digits. In a network with a lot of printers, it would be
advisable for the people setting up the printers to take a
moment to give each one a descriptive name, but in the event
they don't, presenting the users with a list of sequentially
numbered printers is a much more desirable default user
experience than showing a list of raw Ethernet addresses.
Appendix E. Name Encodings in the Domain Name System
Although the original DNS specifications [RFC1033] [RFC1034]
[RFC1035] recommend that host names contain only letters, digits, and
hyphens (because of the limitations of the typing-based user
interfaces of that era), Service Instance Names are not host names.
Users generally access a service by selecting it from a list
presented by a user interface, not by typing in its Service Instance
Name. "Clarifications to the DNS Specification" [RFC2181] directly
discusses the subject of allowable character set in Section 11 ("Name
syntax"), and explicitly states that the traditional letters-digits-
hyphens rule applies only to conventional host names:
Occasionally it is assumed that the Domain Name System serves only
the purpose of mapping Internet host names to data, and mapping
Internet addresses to host names. This is not correct, the DNS is
a general (if somewhat limited) hierarchical database, and can
store almost any kind of data, for almost any purpose.
The DNS itself places only one restriction on the particular
labels that can be used to identify resource records. That one
restriction relates to the length of the label and the full name.
The length of any one label is limited to between 1 and 63 octets.
A full domain name is limited to 255 octets (including the
separators). The zero length full name is defined as representing
the root of the DNS tree, and is typically written and displayed
as ".". Those restrictions aside, any binary string whatever can
be used as the label of any resource record. Similarly, any
binary string can serve as the value of any record that includes a
domain name as some or all of its value (SOA, NS, MX, PTR, CNAME,
and any others that may be added). Implementations of the DNS
protocols must not place any restrictions on the labels that can
be used. In particular, DNS servers must not refuse to serve a
zone because it contains labels that might not be acceptable to
some DNS client programs.
Note that just because DNS-based Service Discovery supports arbitrary
UTF-8-encoded names doesn't mean that any particular user or
administrator is obliged to make use of that capability. Any user is
free, if they wish, to continue naming their services using only
letters, digits, and hyphens, with no spaces, capital letters, or
Appendix F. "Continuous Live Update" Browsing Model
Of particular concern in the design of DNS-SD, especially when used
in conjunction with ad hoc Multicast DNS, is the dynamic nature of
service discovery in a changing network environment. Other service
discovery protocols seem to have been designed with an implicit
unstated assumption that the usage model is:
(a) client software calls the service discovery API,
(b) service discovery code spends a few seconds getting a list of
instances available at a particular moment in time, and then
(c) client software displays the list for the user to select from.
Superficially this usage model seems reasonable, but the problem is
that it's too optimistic. It only considers the success case, where
the software immediately finds the service instance the user is
In the case where the user is looking for (say) a particular printer,
and that printer is not turned on or not connected, the user first
has to attempt to remedy the problem, and then has to click a
"refresh" button to retry the service discovery to find out whether
they were successful. Because nothing happens instantaneously in
networking, and packets can be lost, necessitating some number of
retransmissions, a service discovery search is not instantaneous and
typically takes a few seconds. As a result, a fairly typical user
(a) display an empty window,
(b) display some animation like a searchlight sweeping back and
forth for ten seconds, and then
(c) at the end of the ten-second search, display a static list
showing what was discovered.
Every time the user clicks the "refresh" button they have to endure
another ten-second wait, and every time the discovered list is
finally shown at the end of the ten-second wait, it's already
beginning to get stale and out-of-date the moment it's displayed on
The service discovery user experience that the DNS-SD designers had
in mind has some rather different properties:
1. Displaying the initial list of discovered services should be
effectively instantaneous -- i.e., typically 0.1 seconds, not 10
2. The list of discovered services should not be getting stale and
out-of-date from the moment it's displayed. The list should be
'live' and should continue to update as new services are
discovered. Because of the delays, packet losses, and
retransmissions inherent in networking, it is to be expected that
sometimes, after the initial list is displayed showing the
majority of discovered services, a few remaining stragglers may
continue to trickle in during the subsequent few seconds. Even
after this stable list has been built and displayed, it should
remain 'live' and should continue to update. At any future time,
be it minutes, hours, or even days later, if a new service of the
desired type is discovered, it should be displayed in the list
automatically, without the user having to click a "refresh"
button or take any other explicit action to update the display.
3. With users getting in the habit of leaving service discovery
windows open, and expecting them to show a continuous 'live' view
of current network reality, this gives us an additional
requirement: deletion of stale services. When a service
discovery list shows just a static snapshot at a moment in time,
then the situation is simple: either a service was discovered and
appears in the list, or it was not and does not. However, when
our list is live and updates continuously with the discovery of
new services, then this implies the corollary: when a service
goes away, it needs to *disappear* from the service discovery
list. Otherwise, the service discovery list would simply grow
monotonically over time, accreting stale data, and would require
a periodic "refresh" (or complete dismissal and recreation) to
restore correct display.
4. Another consequence of users leaving service discovery windows
open for extended periods of time is that these windows should
update not only in response to services coming and going, but
also in response to changes in configuration and connectivity of
the client machine itself. For example, if a user opens a
service discovery window when the client machine has no network
connectivity, then the window will typically appear empty, with
no discovered services. When the user connects an Ethernet cable
or joins an 802.11 [IEEEW] wireless network the window should
then automatically populate with discovered services, without
requiring any explicit user action. If the user disconnects the
Ethernet cable or turns off 802.11 wireless then all the services
discovered via that network interface should automatically
disappear. If the user switches from one 802.11 wireless access
point to another, the service discovery window should
automatically update to remove all the services discovered via
the old wireless access point, and add all the services
discovered via the new one.
Appendix G. Deployment History
In July 1997, in an email to the firstname.lastname@example.org
mailing list, Stuart Cheshire first proposed the idea of running the
AppleTalk Name Binding Protocol [RFC6760] over IP. As a result of
this and related IETF discussions, the IETF Zeroconf working group
was chartered September 1999. After various working group
discussions and other informal IETF discussions, several Internet-
Drafts were written that were loosely related to the general themes
of DNS and multicast, but did not address the service discovery
aspect of NBP.
In April 2000, Stuart Cheshire registered IPv4 multicast address
22.214.171.124 with IANA and began writing code to test and develop the
idea of performing NBP-like service discovery using Multicast DNS,
which was documented in a group of three Internet-Drafts:
o "Requirements for a Protocol to Replace the AppleTalk Name Binding
Protocol (NBP)" [RFC6760] is an overview explaining the AppleTalk
Name Binding Protocol, because many in the IETF community had
little first-hand experience using AppleTalk, and confusion in the
IETF community about what AppleTalk NBP did was causing confusion
about what would be required in an IP-based replacement.
o "Discovering Named Instances of Abstract Services using DNS"
[NIAS], which later became this document, proposed a way to
perform NBP-like service discovery using DNS-compatible names and
o "Multicast DNS" [RFC6762] specifies a way to transport those DNS-
compatible queries and responses using IP multicast, for zero-
configuration environments where no conventional Unicast DNS
server was available.
In 2001, an update to Mac OS 9 added resolver library support for
host name lookup using Multicast DNS. If the user typed a name such
as "MyPrinter.local." into any piece of networking software that used
the standard Mac OS 9 name lookup APIs, then those name lookup APIs
would recognize the name as a dot-local name and query for it by
sending simple one-shot Multicast DNS queries to 126.96.36.199:5353.
This enabled the user to, for example, enter the name
"MyPrinter.local." into their web browser in order to view a
printer's status and configuration web page, or enter the name
"MyPrinter.local." into the printer setup utility to create a print
queue for printing documents on that printer.
Multicast DNS responder software, with full service discovery, first
began shipping to end users in volume with the launch of Mac OS X
10.2 "Jaguar" in August 2002, and network printer makers (who had
historically supported AppleTalk in their network printers and were
receptive to IP-based technologies that could offer them similar
ease-of-use) started adopting Multicast DNS shortly thereafter.
In September 2002, Apple released the source code for the
mDNSResponder daemon as Open Source under Apple's standard Apple
Public Source License (APSL).
Multicast DNS responder software became available for Microsoft
Windows users in June 2004 with the launch of Apple's "Rendezvous for
Windows" (now "Bonjour for Windows"), both in executable form (a
downloadable installer for end users) and as Open Source (one of the
supported platforms within Apple's body of cross-platform code in the
publicly accessible mDNSResponder CVS source code repository) [BJ].
In August 2006, Apple re-licensed the cross-platform mDNSResponder
source code under the Apache License, Version 2.0.
In addition to desktop and laptop computers running Mac OS X and
Microsoft Windows, Multicast DNS is now implemented in a wide range
of hardware devices, such as Apple's "AirPort" wireless base
stations, iPhone and iPad, and in home gateways from other vendors,
network printers, network cameras, TiVo DVRs, etc.
The Open Source community has produced many independent
implementations of Multicast DNS, some in C like Apple's
mDNSResponder daemon, and others in a variety of different languages
including Java, Python, Perl, and C#/Mono.
In January 2007, the IETF published the Informational RFC "Link-Local
Multicast Name Resolution (LLMNR)" [RFC4795], which is substantially
similar to Multicast DNS, but incompatible in some small but
important ways. In particular, the LLMNR design explicitly excluded
support for service discovery, which made it an unsuitable candidate
for a protocol to replace AppleTalk NBP [RFC6760].
While the original focus of Multicast DNS and DNS-Based Service
Discovery was for zero-configuration environments without a
conventional Unicast DNS server, DNS-Based Service Discovery also
works using Unicast DNS servers, using DNS Update [RFC2136] [RFC3007]
to create service discovery records and standard DNS queries to query
for them. Apple's Back to My Mac service, launched with Mac OS X
10.5 "Leopard" in October 2007, uses DNS-Based Service Discovery over
Unicast DNS [RFC6281].
In June 2012, Google's Android operating system added native support
for DNS-SD and Multicast DNS with the android.net.nsd.NsdManager
class in Android 4.1 "Jelly Bean" (API Level 16) [NSD].
1 Infinite Loop
Cupertino, CA 95014
Phone: +1 408 974 3207
1 Infinite Loop
Cupertino, CA 95014
Phone: +1 408 974 4368