AppleTalk

After the release of the Apple Lisa computer in January 1983, Apple invested considerable effort in the development of a local area networking (LAN) system for the machines. Known as AppleNet, it was based on the seminal Xerox XNS protocol stack[1] but running on a custom 1 Mbit/s coaxial cable system rather than Xerox's 2.94 Mbit/s Ethernet. AppleNet was announced early in 1983 with a full introduction at the target price of $500 for plug-in AppleNet cards for the Lisa and the Apple II.[2]

At that time, early LAN systems were just coming to market, including Ethernet, Token Ring, Econet, and ARCNET. This was a topic of major commercial effort at the time, dominating shows like the National Computer Conference (NCC) in Anaheim in May 1983. All of the systems were jockeying for position in the market, but even at this time, Ethernet's widespread acceptance suggested it was to become a de facto standard.[3] It was at this show that Steve Jobs asked Gursharan Sidhu a seemingly innocuous question: "Why has networking not caught on?"[4]

Four months later, in October, AppleNet was cancelled. At the time, they announced that "Apple realized that it's not in the business to create a networking system. We built and used AppleNet in-house, but we realized that if we had shipped it, we would have seen new standards coming up."[5] In January, Jobs announced that they would instead be supporting IBM's Token Ring, which he expected to come out in a "few months".[5]

Through this period, Apple was deep in development of the Macintosh computer. During development, engineers had made the decision to use the Zilog 8530 serial controller chip (SCC) instead of the lower-cost and more common UART to provide serial port connections.[6] The SCC cost about $5 more than a UART, but offered much higher speeds of up to 250 kilobits per second (or higher with additional hardware) and internally supported a number of basic networking-like protocols like IBM's Bisync.[7]

The SCC was chosen because it would allow multiple devices to be attached to the port. Peripherals equipped with similar SCCs could communicate using the built-in protocols, interleaving their data with other peripherals on the same bus. This would eliminate the need for more ports on the back of the machine, and allowed for the elimination of expansion slots for supporting more complex devices. The initial concept was known as AppleBus, envisioning a system controlled by the host Macintosh polling "dumb" devices in a fashion similar to the modern Universal Serial Bus.[8]

The Macintosh team had already begun work on what would become the LaserWriter and had considered a number of other options to answer the question of how to share these expensive machines and other resources. A series of memos from Bob Belleville clarified these concepts, outlining the Mac, LaserWriter, and a file server system which would become the Macintosh Office.[4] By late 1983 it was clear that IBM's Token Ring would not be ready in time for the launch of the Mac, and might miss the launch of these other products as well. In the end, Token Ring would not ship until October 1985.[9]

Jobs' earlier question to Sidhu had already sparked a number of ideas. When AppleNet was cancelled in October, Sidhu led an effort to develop a new networking system based on the AppleBus hardware. This new system would not have to conform to any existing preconceptions, and was designed to be worthy of the Mac – a system that was user-installable, had zero configuration, and no fixed network addresses – in short, a true plug-and-play network.[10] Considerable effort was needed, but by the time the Mac was released, the basic concepts had been outlined, and some of the low-level protocols were on their way to completion. Sidhu mentioned the work to Belleville only two hours after the Mac was announced.[4]

The "new" AppleBus was announced in early 1984,[N 1] allowing direct connection from the Mac or Lisa through a small box that is plugged into the serial port and connected via cables to the next computer upstream and downstream. Adaptors for Apple II and Apple III were also announced.[11] Apple also announced that AppleBus networks could be attached to, and would appear to be a single node within, a Token Ring system.[5] Details of how this would work were sketchy.[5]

Just prior to its release in early 1985, AppleBus was renamed AppleTalk. Initially marketed as AppleTalk Personal Network, it comprised a family of network protocols and a physical layer.

The physical layer had a number of limitations, including a speed of only 230.4 kbit/s, a maximum distance of 1,000 feet (300 m) from end to end, and only 32 nodes per LAN.[12] But as the basic hardware was built into the Mac, adding nodes only cost about $50 for the adaptor box. In comparison, Ethernet or Token Ring cards cost hundreds or thousands of dollars. Additionally, the entire networking stack required only about 6 kB of RAM, allowing it to run on any Mac.[13]

The relatively slow speed of AppleTalk allowed further reductions in cost. Instead of using RS-422's balanced transmit and receive circuits, the AppleTalk cabling used a single common electrical ground, which limited speeds to about 500 kbit/s, but allowed one conductor to be removed. This meant that common three-conductor cables could be used for wiring. Additionally, the adaptors were designed to be "self-terminating", meaning that nodes at the end of the network could simply leave their last connector unconnected. There was no need for the wires to be connected back together into a loop, nor the need for hubs or other devices.

The system was designed for future expansion; the addressing system allowed for expansion to 255 nodes in a LAN (although only 32 could be used at that time), and by using "bridges" (which came to be known as "routers", although technically not the same) one could interconnect LANs into larger collections. "Zones" allowed devices to be addressed within a bridge-connected internet. Additionally, AppleTalk was designed from the start to allow use with any potential underlying physical link,[14] and within a few years, the physical layer would be renamed LocalTalk, so as to differentiate it from the AppleTalk protocols.

The main advantage of AppleTalk was that it was completely maintenance-free. To join a device to a network, a user simply plugged the adaptor into the machine, then connected a cable from it to any free port on any other adaptor. The AppleTalk network stack negotiated a network address, assigned the computer a human-readable name, and compiled a list of the names and types of other machines on the network so the user could browse the devices through the Chooser. AppleTalk was so easy to use that ad hoc networks tended to appear whenever multiple Macs were in the same room.[15] Apple would later use this in an advertisement showing a network being created between two seats in an airplane.[16]

A thriving 3rd party market for AppleTalk devices developed over the next few years. One particularly notable example was an alternate adaptor designed by BMUG and commercialized by Farallon as PhoneNET in 1987.[17] This was essentially a replacement for Apple's connector that had conventional phone jacks instead of Apple's round connectors. PhoneNet allowed AppleTalk networks to be connected together using normal telephone wires, and with very little extra work, could run analog phones and AppleTalk on a single four-conductor phone cable.

Other companies took advantage of the SCC's ability to read external clocks in order to support higher transmission speeds, up to 1 Mbit/s. In these systems, the external adaptor also included its own clock, and used that to signal the SCC's clock input pins. The best-known such system was Centram's FlashTalk, which ran at 768 kbit/s, and was intended to be used with their TOPS networking system.[18] A similar solution was the 850 kbit/s DaynaTalk, which used a separate box that plugged in between the computer and a normal LocalTalk/PhoneNet box. Dayna also offered a PC expansion card that ran up to 1.7 Mbit/s when talking to other Dayna PC cards.[19][20] Several other systems also existed with even higher performance, but these often required special cabling that was incompatible with LocalTalk/PhoneNet, and also required patches to the networking stack that often caused problems.

As Apple expanded into more commercial and education markets, they needed to integrate AppleTalk into existing network installations. Many of these organizations had already invested in a very expensive Ethernet infrastructure and there was no direct way to connect a Macintosh to Ethernet. AppleTalk included a protocol structure for interconnecting AppleTalk subnets and so as a solution, EtherTalk was initially created to use the Ethernet as a backbone between LocalTalk subnets. To accomplish this, organizations would need to purchase a LocalTalk-to-Ethernet bridge and Apple left it to third parties to produce these products.[21] A number of companies responded, including Hayes and a few newly formed companies like Kinetics.

By 1987, Ethernet was clearly winning the standards battle over Token Ring, and in the middle of that year, Apple introduced EtherTalk 1.0, an implementation of the AppleTalk protocol over the Ethernet physical layer. Introduced for the newly released Macintosh II computer, Apple's first Macintosh with expansion slots, the operating system included a new Network control panel that allowed the user to select which physical connection to use for networking (from "Built-in" or "EtherTalk"). At introduction, Ethernet interface cards were available from 3Com and Kinetics that plugged into a Nubus slot in the machine. The new networking stack also expanded the system to allow a full 255 nodes per LAN. With EtherTalk's release, AppleTalk Personal Network was renamed LocalTalk,[22] the name it would be known under for the bulk of its life. Token Ring would later be supported with a similar TokenTalk product, which used the same Network control panel and underlying software. Over time, many third-party companies would introduce compatible Ethernet and Token Ring cards that used these same drivers.

The appearance of a Macintosh with a direct Ethernet connection also magnified the Ethernet and LocalTalk compatibility problem: Networks with new and old Macs needed some way to communicate with each other. This could be as simple as a network of Ethernet Mac II's trying to talk to a LaserWriter that only connected to LocalTalk. Apple initially relied on the aforementioned LocalTalk-to-Ethernet bridge products, but contrary to Apple's belief that these would be low-volume products, by the end of 1987, 130,000 such networks were in use. AppleTalk was at that time the most used networking system in the world, with over three times the installations of any other vendor.[23]

1987 also marked the introduction of the AppleShare product, a dedicated file server that ran on any Mac with 512 kB of RAM or more. A common AppleShare machine was the Mac Plus with an external SCSI hard drive. AppleShare was the #3 network operating system in the late 1980s, behind Novell NetWare and Microsoft's MS-Net.[24] AppleShare was effectively the replacement for the failed Macintosh Office efforts, which had been based on a dedicated file server device.

A significant re-design was released in 1989 as AppleTalk Phase II. In many ways, Phase II can be considered an effort to make the earlier version (never called Phase I) more generic. LANs could now support more than 255 nodes, and zones were no longer associated with physical networks but were entirely virtual constructs used simply to organize nodes. For instance, one could now make a "Printers" zone that would list all the printers in an organization, or one might want to place that same device in the "2nd Floor" zone to indicate its physical location. Phase II also included changes to the underlying inter-networking protocols to make them less "chatty", which had previously been a serious problem on networks that bridged over wide-area networks.[25]

By this point, Apple had a wide variety of communications products under development, and many of these were announced along with AppleTalk Phase II. These included updates to EtherTalk and TokenTalk, AppleTalk software and LocalTalk hardware for the IBM PC, EtherTalk for Apple's A/UX operating system allowing it to use LaserPrinters and other network resources, and the Mac X.25 and MacX products.

Ethernet had become almost universal by 1990, and it was time to build Ethernet into Macs direct from the factory. However, the physical wiring used by these networks was not yet completely standardized. Apple solved this problem using a single port on the back of the computer into which the user could plug an adaptor for any given cabling system. This FriendlyNet system was based on the industry-standard Attachment Unit Interface or AUI, but deliberately chose a non-standard connector that was smaller and easier to use, which they called "Apple AUI", or AAUI. FriendlyNet was first introduced on the Quadra 700 and Quadra 900 computers, and used across much of the Mac line for some time.[26] As with LocalTalk, a number of 3rd party FriendlyNet adaptors quickly appeared.

As 10BASE-T became the de facto cabling system for Ethernet, second-generation Power Macintosh machines added a 10BASE-T port in addition to AAUI. The PowerBook 3400c and lower-end Power Macs also added 10BASE-T. The Power Macintosh 7300/8600/9600 were the final Macs to include AAUI, and 10BASE-T became universal starting with the Power Macintosh G3 and PowerBook G3.

From the beginning of AppleTalk, users wanted to connect the Macintosh to the TCP/IP network environments. In 1984, Bill Croft at Stanford University pioneered the development of IP packets encapsulated in DDP as part of the SEAGATE (Stanford Ethernet–AppleTalk Gateway) project. SEAGATE was commercialized by Kinetics in their LocalTalk-to-Ethernet bridge as an additional routing option. A few years later, MacIP, was separated from the SEAGATE code and became the de facto method for IP packets to be routed over LocalTalk networks. By 1986, Columbia University released the first version of the Columbia AppleTalk Package (CAP) that allowed higher integration of Unix, TCP/IP, and AppleTalk environments. In 1988, Apple released MacTCP, a system that allowed the Mac to support TCP/IP on machines with suitable Ethernet hardware. However, this left many universities with the problem of supporting IP on their many LocalTalk-equipped Macs. It was soon common to include MacIP support in LocalTalk-to-Ethernet bridges.[26] MacTCP would not become a standard part of the Classic Mac OS until 1994,[27] by which time it also supported SNMP and PPP.

For some time in the early 1990s, the Mac was a primary client on the rapidly expanding Internet. Among the better-known programs in wide use were Fetch, Eudora, eXodus, NewsWatcher, and the NCSA packages, especially NCSA Mosaic[28] and its offspring, Netscape Navigator.[29] Additionally, a number of server products appeared that allowed the Mac to host Internet content. Through this period, Macs had about 2 to 3 times as many clients connected to the Internet as any other platform,[30] despite the relatively small overall microcomputer market share.

As the world quickly moved to IP for both LAN and WAN uses, Apple was faced with maintaining two increasingly outdated code bases on an ever-wider group of machines as well as the introduction of the PowerPC based machines. This led to the Open Transport efforts, which re-implemented both MacTCP and AppleTalk on an entirely new code base adapted from the Unix standard STREAMS. Early versions had problems and did not become stable for some time.[31] By that point, Apple was deep in their ultimately doomed Copland efforts.

With the purchase of NeXT and subsequent development of Mac OS X, AppleTalk was strictly a legacy system. Support was added to OS X in order to provide support for a large number of existing AppleTalk devices, notably laser printers and file shares, but alternate connection solutions common in this era, notably USB for printers, limited their demand. As Apple abandoned many of these product categories, and all new systems were based on IP, AppleTalk became less and less common. AppleTalk support was finally removed from the MacOS in Mac OS X v10.6 in 2009.[32]

However, the loss of AppleTalk did not reduce the desire for networking solutions that combined its ease of use with IP routing. Apple has led the development of many such efforts, from the introduction of the AirPort router to the development of the Zero-configuration networking system and their implementation of it, Bonjour.

As of 2020, AppleTalk support has been completely removed from legacy support with macOS 11 Big Sur.

AARP resolves AppleTalk addresses to link layer addresses.[36] It is functionally equivalent to ARP and obtains address resolution by a method very similar to ARP.

AARP is a fairly simple system. When powered on, an AppleTalk machine broadcasts an AARP probe packet asking for a network address, intending to hear back from controllers such as routers. If no address is provided, one is picked at random from the "base subnet", 0. It then broadcasts another packet saying "I am selecting this address", and then waits to see if anyone else on the network complains. If another machine has that address, it will pick another address, and keep trying until it finds a free one.[37] On a network with many machines it may take several tries before a free address is found, so for performance purposes the successful address is "written down" in NVRAM and used as the default address in the future. This means that in most real-world setups where machines are added a few at a time, only one or two tries are needed before the address effectively become constant.

This was a comparatively late addition to the AppleTalk protocol suite, done when it became clear that a TCP-style reliable connection-oriented transport was needed. Significant differences from TCP were that:

• a connection attempt could be rejected
• there were no "half-open" connections; once one end initiated a tear-down of the connection, the whole connection would be closed (i.e., ADSP is full-duplex, not dual simplex).
• AppleTalk had an included attention message system which allowed short messages to be sent which would bypass the normal stream data flow. These were delivered reliably but out of order with respect to the stream. Any attention message would be delivered as soon as possible instead of waiting for the current stream byte sequence point to become current.[38]

The Apple Filing Protocol (AFP), formerly AppleTalk Filing Protocol, is the protocol for communicating with AppleShare file servers. Built on top of AppleTalk Session Protocol (for legacy AFP over DDP) or the Data Stream Interface (for AFP over TCP), it provides services for authenticating users (extensible to different authentication methods including two-way random-number exchange) and for performing operations specific to the Macintosh HFS filesystem. AFP is still in use in macOS, even though most other AppleTalk protocols have been deprecated.

ASP was an intermediate protocol, built on top of ATP, which in turn was the foundation of AFP. It provided basic services for requesting responses to arbitrary commands d performing out-of-band status queries. It also allowed the server to send asynchronous attention messages to the client.

DDP was the lowest-level data-link-independent transport protocol. It provided a datagram service with no guarantees of delivery. All application-level protocols, including the infrastructure protocols NBP, RTMP and ZIP, were built on top of DDP. AppleTalk's DDP corresponds closely to the Network layer of the Open Systems Interconnection (OSI) communication model.

Name Binding Protocol was a dynamic, distributed system for managing AppleTalk names. When a service started up on a machine, it registered a name for itself as chosen by a human administrator. At this point, NBP provided a system for checking that no other machine had already registered the same name. Later, when a client wanted to access that service, it used NBP to query machines to find that service. NBP provided browsability ("what are the names of all the services available?") as well as the ability to find a service with a particular name.[36] Names were human readable, containing spaces, upper and lower case letters, and including support for searching.

AEP (AppleTalk Echo Protocol) is a transport layer protocol designed to test the reachability of network nodes.[36] AEP generates packets to be sent to the network node and is identified in the Type field of a packet as an AEP packet. The packet is first passed to the source DDP. After it is identified as an AEP packet, it is forwarded to the node where the packet is examined by the DDP at the destination. After the packet is identified as an AEP packet, the packet is then copied and a field in the packet is altered to create an AEP reply packet, and is then returned to the source node.

PAP was the standard way of communicating with PostScript printers. It was built on top of ATP.[36] When a PAP connection was opened, each end sent the other an ATP request which basically meant "send me more data". The client's response to the server was to send a block of PostScript code, while the server could respond with any diagnostic messages that might be generated as a result, after which another "send-more-data" request was sent. This use of ATP provided automatic flow control; each end could only send data to the other end if there was an outstanding ATP request to respond to.

PAP also provided for out-of-band status queries, handled by separate ATP transactions. Even while it was busy servicing a print job from one client, a PAP server could continue to respond to status requests from any number of other clients. This allowed other Macintoshes on the LAN that were waiting to print to display status messages indicating that the printer was busy, and what the job was that it was busy with.

RTMP was the protocol by which routers kept each other informed about the topology of the network.[36] This was the only part of AppleTalk that required periodic unsolicited broadcasts: every 10 seconds, each router had to send out a list of all the network numbers it knew about and how far away it thought they were.

ZIP was the protocol by which AppleTalk network numbers were associated with zone names.[36] A zone was a subdivision of the network that made sense to humans (for example, "Accounting Department"); but while a network number had to be assigned to a topologically-contiguous section of the network, a zone could include several different discontiguous portions of the network.