A growing number of embedded devices are connected to TCP/IP networks. And many of these devices are no longer passive network nodes, waiting for someone else to control them. They are full-blown network citizens, actively communicating with their peers and often relying on the network services provided by other devices to do their job. For this to work, all these devices must be configured properly. Configuring the network parameters (e.g., IP address, netmask, etc.) of a device is a tedious task, as many devices do not have an appropriate user interface to do this comfortably. This is especially an issue with consumer devices, where the user might not even have the necessary technical knowledge to configure such a device. And as the number of devices in a network grows, configuring each device separately is no longer practical. There from comes the need for the automatic configuration of network devices and the automatic discovery of network services. In recent years, the industry has come up with a variety of different technologies and specifications to address this.
Generally, there are three fundamental issues that must be dealt with for the automatic discovery of network devices and services, and three more issues that must be dealt with for the actual invocation or use of the discovered services.
Every device must be assigned a unique network address. In case of TCP/IP networks this can be done with the help of a DHCP (Dynamic Host Configuration Protocol) server. Should, however, no DHCP server be available (for example, in typical home user networks), another way of assigning an IP addresses must be found. Apart from manual configuration, which is often undesirable, a method called APIPA (Automatic Private IP Addressing, or AutoIP for short) is used. In this case, a device's TCP/IP stack randomly chooses an IP address in the private (link-local) IP address range 169.254.0.0 to 169.254.255.255. To prevent two or more devices from accidentally selecting the same address, each device must probe, using the ARP (Address Resolution Protocol), whether the chosen address is available.
Whenever a user (or device) wants to access another device over the network, he usually wants to refer to the device by its name, not by its IP address. In a typical TCP/IP network, a DNS (Domain Name System) server is used to map domain names to IP addresses. Again, if no DNS server is available, such as in a typical home network, another way of resolving names to IP addresses is required. Multicast DNS (mDNS), as used by Zeroconf, is such an alternative technology.
A user or device must be able to find a service provided by one or more devices in the network. The important part here is that the user (or device) usually does not care which device implements the service, as long as the service with specific properties is available and accessible. A typical example for service discovery is: I need to print a document in color. Give me an IP address and port number where I can send the print job to, using the Internet Printing Protocol (IPP), so that the document will be printed in color.
What all technologies for service discovery have in common is, that they make use of IP multicasting. IP multicasting uses addresses in a special address range (126.96.36.199 to 188.8.131.52). A packet (typically, a UDP packet) sent to a multicast address is received by all hosts listening to that specific address.
Service discovery is implemented in the following way:
- An application or device that needs a certain service sends a request describing the required properties of the service to a specific multicast address (and port number).
- Other applications or devices on the same network receive the request, and if they provide the requested service themselves (or know another device that implements the service), respond with a message describing where the service can be found.
- The application or device searching for the service collects all responses, and from the responses chooses the one service provider it is going to use.
In addition, devices that join or leave a network can send announcements to other devices describing the availability of the services they provide.
Once a certain service has been discovered, it may be necessary to obtain more information about the service. For example, if a service consists of multiple operations, it is necessary to find out exactly what operations are supported, and what arguments must be passed to them. This is the purpose of service description.
In case only well-defined service protocols are used (e.g., HTTP, Internet Printing, or media streaming), service description is not necessary, because the only information needed to access the service is the network address (IP address and port number, or URI), and this information can be obtained by service discovery. In other cases, the information obtained via service discovery is insufficient to successfully access the service. In such a case, service discovery only returns the address of a network resource that provides detailed information about the capabilities of the service, and how to access them.
After an application has obtained enough information about the services it wants to access — either via service discovery alone, or together with service description, the application will access or invoke them. This is usually beyond the scope of most service discovery technologies, and the domain of specialized technologies and protocols. Examples for such technologies are HTTP (HyperText Transfer Protocol), SOAP, Java RMI (Remote Method Invocation), CORBA (Common Object Request Broker Architecture), or media streaming protocols such as RTSP (Real Time Streaming Protocol).
Some technologies (UPnP and Jini) provide facilities to present a device specific user interface to the user, via another device (e.g., a central management workstation, the user's PC or TV set). Such a user interface can be used to allow the user to directly interact with a device, in order to configure it, or to use some of its functions not available otherwise.
This requires that the user interface is implemented in a device independent way. One way to do this is to implement the user interface in HTML (HyperText Markup Language), served by an embedded web server built into the device. Another way is to implement the user interface in Java, so that it can be run everywhere a Java Virtual Machine is available. The user interface code than talks to the device over a network connection, using a possibly proprietary protocol.
Zero Configuration Networking (Zeroconf) is a technology originally developed by Apple and promoted under the trademark name Bonjour. Zeroconf is built upon technologies known as multicast DNS (mDNS) and DNS Service Discovery (DNS-SD), which themselves are based on the proven DNS protocol. Its most popular uses are in network printers, network cameras and for sharing music using Apple's iTunes music jukebox software. Open source implementations are available from Apple and others for all important platforms.
Zeroconf uses DHCP and AutoIP for address assignment if no DHCP server is available. A special variant of the well-known DNS protocol, multicast DNS, is used for name resolution in case no DNS server is available. In mDNS, a name resolution query is sent not directly to a DNS server, as with traditional DNS, but to a multicast address. All network devices supporting Zeroconf and listening to this multicast address respond to name resolution queries, if they have the requested information.
Finally, for service discovery, an extension of the DNS protocol called DNS Service Discovery is used. DNS-SD can be used both with multicast DNS (the usual way), or with traditional unicast queries to a DNS server. With the additional support for DNS Update and DNS Long Lived Queries, two extensions of the DNS protocol, Zeroconf can be used to announce services across the global internet. In this case, an ordinary DNS server is used to make the service information available. Periodic automatic updates of the DNS server's database ensure that its service information is up to date.
A major advantage of Zeroconf is that it is based on proven technology. Also, it can be implemented in a very resource efficient way, making it a good choice for resource constrained embedded devices.
Beside heavy use by Apple in many of its applications for Mac OS X and Windows (iTunes), Zeroconf is popular among manufacturers of printers and network cameras.
Two implementations of Zeroconf are currently in widespread use.
Apple's Bonjour was the first implementation of Zeroconf. The implementation is available under an open source license and can be used on different platforms like Mac OS X, Windows, Linux or VxWorks. Bonjour is an integrated part of Mac OS X and iOS. A Windows installer (and SDK) is available as well.
Avahi is another open source implementation of Zeroconf, targeted mostly at Linux, although it can also be used on other POSIX platforms like FreeBSD or Solaris. Avahi has been integrated into many Linux distributions such as Debian, Ubuntu, SuSE and others.
The DNS-SD library from Applied Informatics provides an easy-to-use and unified programming interface for integrating Zeroconf features (service discovery and host name resolution) into a C++ application. The Applied Informatics DNS-SD library does not implement its own mDNS and DNS-SD protocol stacks, but rather uses an existing Zeroconf implementation for that purpose. Apple's Bonjour and Avahi can be used as backend for the DNS-SD library.
A great advantage of the library is that it provides a unified programming interface. For the programmer, it's completely transparent whether Avahi or Bonjour is used as backend.