ARCNET ARCNET is one of the oldest existing LAN systems, originally developed in 1977 by Datapoint. The primary architect of ARCNET was John Murphy, who is now a software engineer with Performance Technology, a developer of LAN operating software. Standard Microsystems Corporation (SMC) developed the first ARCNET controller chip in 1981, and the first ARCNET controller board for IBM type PCs in 1983. Since then ARCNET has become one of the most popular LAN hardware systems in the world, comprising an estimated 25% of new LAN installations. This figure includes all LAN types; the estimate for PC LANs is much higher. For example, it is estimated that more than half the installations of Novell's popular NetWare operating software are using ARCNET hardware. According to Ben Wolfe, chairman of the ARCNET Trade Association, there are approximately 135 manufacturers of ARCNET products. ARCNET is supported by many LAN operating systems, including Novell NetWare, Banyan Vines, Western Digital Vianet, CBIS Network OS, Performance Technologies POWERLAN, Univation LifeNET and many others. ARCNET is also used widely in industrial and manufacturing control systems. ARCNET interfaces are available for various expansion buses, including IBM PC, PC AT, PS/2, S-100, Multibus, VME bus, Qbus, STD bus STE Bus and Nubus (used by the Apple Macintosh). Mr. Wolfe stated that more than 1,300,000 ARCNET nodes will be installed by the end of 1988. The major manufacturers of ARCNET currently include SMC, Pure Data Corporation, Thomas Conrad Corporation and Tiara Computer Systems. Why is a LAN system that does not bear an IEEE stamp of approval so popular? There are several important reasons: reliability, flexibility, performance, ease of installation and ease of fault diagnosis are among them. The consensus among LAN resellers and installers is that ARCNET is the easiest LAN to install and troubleshoot. Once installed, ARCNET controllers rarely cause trouble, and, due to the "star-cluster" topology of ARCNET, isolating sections of the network for diagnosis is quite easy. While ARCNET at 2.5 megabits per second is not the fastest LAN around, due to its token-passing protocol it provides reasonable performance that does not degrade quickly under load. ARCNET also provides a great deal of cabling flexibility. Additionally, virtually any manufacturer's ARCNET controller will function properly in the same network with any other manufacturer's controller. ARCNET seems to have achieved a degree of standardization not always available in the 802.3 Ethernet and Starlan world. Topology and cabling The standard topology for ARCNET is a star-cluster scheme, with ARCNET nodes (or stations) attached to distribution devices called hubs. There are two types of ARCNET hubs: active and passive. An active hub is a powered unit that acts as a distribution device and signal amplifier, while a passive hub only distributes signals over short distances. Active hubs may have between four and 64 ports, while passive hubs have 4 ports. By using multiple hubs, an ARCNET can have a diameter as large as 20,000 feet, meaning a maximum cabling distance of 20,000 feet from any node on the network to any other node. Standard ARCNET cable is RG-62 coaxial cable with an impedance of 93 ohms. Enhancements made over the last few years have added cabling options (more about this later). Cable Access Scheme ARCNET is a Token Bus network. A 'bus' network has a common cable set, with all signals being broadcast across the entire network simultaneously. The signals are only received and acknowledged by the stations they are addressed to. ARCNET uses a Logical Token Passing scheme to control cable access. In token passing systems, there is an electronic signal called a "token" that is passed from station to station in a pre-determined order. If a station has possession of the token, it can send a packet of information. If it does not have possession of the token, it must wait until the token is passed to it by the previous station before it can transmit. Basically, token passing systems are "contention-free" systems. Unlike the contention access scheme of Ethernet and others (called CSMA, for Carrier Sense Multiple Access, where any station can transmit if the cable is free. When two stations transmit simultaneously, a "data collision" occurs, and data packets must be re-transmitted by both stations.), each station "waits its turn" until it receives the token. In token passing systems network access is guaranteed, while access is random in contention systems. Performance in heavy traffic situations is generally better with token passing systems than with contention systems. Token passing functions differently in ring systems, such as the IBM Token_Ring, than it does in bus systems, such as ARCNET. In a token ring system, the token is physically passed from one station in the ring to the next station in the ring. It is then regenerated and passed to the following station. In a token bus, such as ARCNET, the token acts as permission to transmit. In simplest terms, each ARCNET station has the address of one other station on the network in its Next ID (NID) register. When a station has possession of the token, it can elect to send a data packet to another station. After it sends a packet (and receives acknowledgement) or if it elects not to send a packet, it broadcasts the token across the network to the station whose address is in its NID register. The addressed station acknowledges receipt of the token and begins the process over again. ARCNET dynamically and automatically reconfigures every time a station is added to or removed from the network. When a station is added, or 'powered on', it transmits a signal that is unambiguously an invalid data pattern and longer than any possible data packet. This "recon burst" interferes with all communication and prevents any node from receiving the token, forcing a reconfiguration. At this point each station sets its NID register to its own address, then waits up to (146 X (255-ID)) microseconds. Since ARCNET addresses are switch selectable in a range of 1 to 255, this means that the station with highest address starts the reconfiguration. It sends out a signal then waits a specific amount of time (usually approximately 75 microseconds) for a response. If it does not receive a response it increments the address in the NID register (in this case its own address), and starts over. After reaching 255 it increments the address to 0 and continues the process until it finds the lowest addressed station on the network. It then puts that station's address in its address buffer and passes it the token. Each node then reconfigures in order. In a worst-case scenario, this process takes a maximum of 61 milliseconds. When a station is removed from the network the process is much simpler. The station with the address just below the removed station attempts to pass the token and waits for a response. If it does not receive a response it increments the value in its NID register by one, sends out a signal and waits for a response. It repeats this procedure until the next station is found, then passes the token to it. Diagnosis Fault diagnosis on ARCNET is usually quite easy. Because of the star cluster topology, it is quite easy to disconnect whole sections of the network for fault isolation. ARCNET also has a built-in diagnostic tool -- the "recon burst". In a healthy network, recon bursts should only occur when stations are being added to the LAN (software, such as the Novell NetWare shell, can also force a recon burst). A hardware problem, however, can often cause spurious reconfigurations. There are several diagnostic software products that, among other things, monitor reconfigurations. These include ARCMonitor from Brightwork Development, Inc., ARCDIAG from Microlan Systems, Inc. and ARCView from SMC. By disconnecting sections of the network and monitoring recon bursts most ARCNET hardware problems can be located. Extended ARCNET Many extensions have been added to ARCNET since its original development. Some of these are: 1. Fiber optics. The maximum distance between standard active ARCNET devices (ARCNET controllers and active hubs) is 2000'. With fiber optics that distance can be extended. With single fiber optic controllers and links that distance can be extended to 4000', while dual fiber systems can extend the distance to 11,500 feet. 2. Linear Bus ARCNET. Another relatively recent development is the Linear Bus ARCNET card, originally developed by SMC and now produced by many manufacturers. This card uses special circuitry to allow up to 8 cards to share the same linear cable. This cable can have a length of up to 1000 feet. Attachment is made using the same type of BNC "T" connector used with thin Ethernet cable. One end of this cable may be connected to an active hub, allowing interconnection with the star-cluster topology, and multiple linear buses may be connected using 2 port "active links". 3. Twisted-Pair ARCNET. ARCNET boards and hubs designed to use unshielded twisted-pair (UTP) phone wire have been developed by SMC (and, again, now manufactured by others). These twisted-pair boards may be cabled (up to 400 feet) to special twisted-pair active hubs and up to 10 boards may be daisy-chained (in Starlan fashion). Unlike other UTP LAN implementations, only a single wire pair is required. 4. 16 Bit AT and Microchannel Interfaces. First released by Thomas Conrad Corp. (16 bit AT) and Pure Data (microchannel), similar controllers are provided by other major ARCNET manufacturers. These boards generally provide better performance due to the 16 bit data path and improved packet buffering. 5. Nodal Priority. SMC now manufactures a 16 bit board designed especially for LAN servers. This board, the PC500FS, can send multiple packets each time it has possession of the token. This means that it can respond to pending requests from multiple work stations on the same round trip of the token. The future of ARCNET While many of the enhancements described above contribute to increased performance, ARCNET is still crippled by two factors: its relatively slow (by today's standards) 2.5 megabit transmission speed, and its inefficient 508 byte packet size. Several groups are attempting to correct the situation. SMC is developing a 5 megabit version of ARCNET that will reportedly have a 2K packet size. Unfortunately, this system will not be compatible with current ARCNET products. Another approach is being taken by the ARCNET Trade Association. They are currently setting the specifications for an enhanced ARCNET that will initially provide a 20 megabit per second transfer rate, and a larger packet size, probably 2048 bytes, possibly 4096. The advantage of this specification is that it will be supported on existing coax cabling (and possibly UTP) and will maintain full downward compatibility with current ARCNET products. The design will allow 20 megabit boards to communicate with 2.5 megabit boards at the slower speed. While the current specification calls for 20 megabits per second, faster speeds are planned in the future, up to 100 megabits per second. Acknowledgements Many thanks to Mike Fischer of Datapoint Corp., John Murphy, formerly of Datapoint, now with Performance Technology, Ben Wolfe of the ARCNET Trade Association and many others for their help with this article. ARCNET is a registered trademark of Datapoint Corp. ARCNET Trade Association 3413 North Kennicott Avenue Suite B Arlington Heights, IL 60004 Datapoint Corporation 9725 Datapoint Drive San Antonio, TX 78284 512-699-7054 Standard Microsystems Corporation 35 Marcus Boulevard Hauppauge, NY 11788 516-273-3100 Performance Technology 800 Lincoln Center 7800 IH 10 West San Antonio, TX 78230 512-349-2000 Pure Data 200 West Beaver Creek Road Richmond Hill, Ontario Canada L4B 1B4 416-731-6444 Thomas Conrad Corporation 8403 Cross Park Drive, #1C Austin, TX 78754 512-836-1935 Tiara Computer Systems 2700 Garcia Avenue Mountain View, CA 94043 415-965-1700 (c) Copyright 1988 Patrick H. Corrigan Commercial reproduction prohibited without written permission.