January 1998

NETWORK

Businesses thrive on communications. Getting information to where it's needed can spell the difference between success and failure. Telephones are good for voice and crude faxed images.

Plants are nature's prototypes for computer networks. The shortest possible distance between leaves and roots is a direct connection. This lets water and food move as fast as possible throughout a plant. A lot of plants do this. They grow fast, but stay small because a lot of resources are needed to give each leaf a direct connection to the roots.

Other plants grow a single stalk to which all the leaves are connected. This saves resources because the tree only needs one trunk. However, the leaves at the top of the tree don't get as much water as those at the bottom nor can they contribute as much food to the rest of the plant.

Trees use a combination of these strategies. A tree uses a large trunk and heavy branches to save resources. At the ends of the branches, leaves are clustered, each with their own stem to shorten their distance to the roots and to each other.

When putting together a computer network, the network designer has similar requirements. For maximum speed the best

networking model is to connect all the computers to the server with individual direct lines. In a large organization this is not practical. Giving everyone a direct connection to the server is complicated and expensive. First, there is the logistical problem of running all the wire. Second, the server has to directly support thousands of computers.

Network designers solve this problem by combining local area networks (LANs) to build a wide area network (WAN). A handful of computers are linked together to form a workgroup. Several workgroups at a single installation form a LAN. Connect together several widely dispersed LANs and you have a WAN. WANs bind smaller networks and single computers together to form larger networks.

Network Technologies

The most common hardware technologies for LANs are Ethernet and Token Ring. Token ring was developed in 1969 and commercialized in 1985. Computers are arranged in a circle. The server sends a signal called a token around the ring. Each computer on the ring can look at the token. If a computer needs to send information, it seizes the token, adds its data and a header describing the destination and retransmits the token on the network.

Token rings pass information at up to 16 megabits per second (Mbps). They are easy to maintain and don't get overloaded easily. Since Token Ring technology has been around since the sixties, it is used in a lot of mainframes and legacy hardware.

Ethernet was developed by Xerox (Palo Alto, CA 415-813-6800) in 1976. It's the most widespread networking hardware standard among personal computers. The original Ethernet used a thick coaxial cable called 10Base-5.

Each of the computers in the network used a transceiver to connect to this cable. Newer versions of Ethernet use either a thin coaxial cable called 10Base-2 or unshielded twisted pair cable called 10Base-T. With 10Base-2 cable, the computers are daisy-chained to each other. In 10Base-T networks, all the computers are connected to a hub.

Because Ethernet is so popular, there are more Ethernet devices than any other network product. In addition to computers, you can get network printers, scanners, RAID subsystems, hubs, ISDN routers and bridges. Ethernet's popularity has led to a tremendous variety of new technologies.

Fast Ethernet lets Ethernet networks send data between computers at up to 100 megabits per second, 10 times faster than traditional Ethernet. Gigabit Ethernet, still under development, is 10 times faster then fast Ethernet. This could become the networking standard for wide area networks.

Token Ring and Ethernet networks are built on dedicated connections. What do you do if you're traveling or working from home? Most people use modems over ordinary telephone lines. The network protocols for modems are Serial Line Internet Protocol (SLIP) and Point to Point Protocol (PPP).

SLIP and PPP were originally designed so modem users could connect to the Internet. SLIP is still solely an Internet protocol. PPP is newer and more flexible. It can carry other high level protocols like Appletalk and Novell. Although SLIP and PPP are normally implemented in software, their low-level nature means they have more in common with Ethernet and Token Ring than with high-level protocols.

SLIP is an extremely simple protocol. SLIP defines framing characters for TCP/IP packets. TCP/IP is the Internet networking protocol. PPP is a much more sophisticated way of connecting over telephone lines.

In addition to start and stop characters, PPP specifies the high-level protocol, a checksum for error correction and negotiates network ID numbers, the number identifying your computer on a network. PPP works with the Internet and other networks. WANs connect multiple local area networks together. Sometimes connected WANs are spread around the world. Sometimes they are in the same building. Either way, WANs carry much more traffic than LANs and are faster. WAN connections have more bandwidth.

The oldest WAN technology is Frame Relay. Frame Relay runs on digital telephone lines. Frame Relay was originally designed for the Internet. It has built in redundancy. Once a packet enters into a Frame Relay network it follows the best available path to the receiving computer. Not all individual packets follow the same path to their destination. Nor do they arrive in the same order they were sent. The computer on the receiving end has to reassemble the packets. Frame Relay uses bandwidth very efficiently. If one network connection is congested, packets are guided through an underutilized connection.

Asynchronous Transfer Mode (ATM) is a newer system. It's the system of choice among large network providers. ATM sets a fixed route between the computers sending and receiving information. This connection stays open for as long as the two computers communicate with each other. ATM is good for streaming information like audio and video. It makes it easier for network providers to bill network usage. ATM makes it easy for network providers to guarantee a minimum quality of service.

A disadvantage of ATM is its small packet size. For ATM to work, all packets sent across a network connection have to be uniform in size. The current standard says all packets are 42 bytes long. Since each packet has a header identifying it and telling it where it should go (five bytes), 12% of the bandwidth is wasted.

The latest network vaporware is Gigabit Ethernet, although pre-standard hardware is available now. Gigabit Ethernet is an extension of traditional Ethernet technology. The original Ethernet supported 10-100 megabits per second. Gigabit Ethernet transfers data at up to 1,000/Mbits per second -- 40% faster than ATM which maxes out at 622/Mbits per second.

Gigabit Ethernet should cost less than equivalent ATM equipment and offer higher performance. Gigabit Ethernet should also work better with installed local area networks, the majority of which are already Ethernet. Gigabit Ethernet's speed makes it an attractive alternative to ATM for high capacity Internet backbones. The low cost may let more companies build intranets. All this is theoretical since Gigabit Ethernet isn't yet ready for volume production.

Network Protocols

Probably the most confusing part of network technology is the difference between network protocols and the underlying network hardware. This is even more confusing because network hardware includes low-level software. Practically all network technology is based on Open Standards Interconnection (OSI). OSI defines the network hardware and software in terms of layers that can be added and swapped as needed.

A network hardware standard is specified by the physical hardware, network interface cards, wire and low-level software or firmware built into the networking hardware. The Ethernet standard states how Ethernet cards should work, the characteristics of the wire used to connect machines together and the size of the packets transmitted.

At least part of how Ethernet works is encoded in low-level software that's loaded into a computer with the operating system or recorded in ROMs built into the Ethernet hardware. This low-level software is similar to the basic input/output software built into the ROMs on a computer motherboard.

The network protocol is the high-level part of a networking system. You can tell the network protocol from the underlying hardware because network protocols are transportable. The network protocol of the Internet, TCP/IP, can be run on just about any underlying hardware.

You can run TCP/IP on Ethernet, Token Ring, ATM, frame relay or PPP. Each of these underlying hardware standards has its own electrical characteristics, packet sizes, etc. TCP/IP specifies its own packet sizes which are then mapped onto the underlying network hardware. If you don't want to use TCP/IP, you could use another network protocol like Novell's IPX or Appletalk on the same hardware.

TCP/IP is the network protocol of the largest network in the world, the Internet. This is a good choice if you're building a large network. Many organizations with large networks build intranets based on TCP/IP. Scalability has a price. TCP/IP requires more network administration than other networks. This makes it less desirable for small LANs. On the plus side, TCP/IP is built into every computer's operating system. TCP/ IP makes it easy to get all the computers on your network on the Internet.

TCP/IP is really two protocols in one. Internet Protocol (IP) was developed first. IP groups data into chunks called packets. Each packet has an address for identification, and a Time To Live (TTL) field that tells the network how long to keep the packet if it can't be delivered. IP packets can be broken up. IP reassembles the packets. This allows IP packets to be transported on a wide variety of network connections.

The other half of TCP/IP is Transmission Control Protocol (TCP). TCP adds the information to IP packets that IP leaves out: error correction, sequence numbers, buffering information and network services. Sequence numbers tell the receiving computer whether the packets have arrived in order and if not, what order the packets should be arranged in.

Because the packets can arrive out of order, the receiver must store them in a buffer. TCP tells the receiver how large the buffer should be. Other network services include elapsed time and time to arrival. Networks can use this information to check usage and to optimize the routes packets take on their journey.

The most popular networking protocol, especially for LANs, is Novell's (Provo, UT 801-861-7000) Netware. Like TCP/IP, Netware is a network protocol that runs on any type of networking hardware. Like TCP/IP, Netware uses OSI's layer model of networking.

Netware is a combination of two protocols, Internetwork Packet Exchange (IPX) and Sequenced Packet Exchange (SPX). Like IP, IPX defines identification numbers for different computers on a network and gives each packet an unique identifier. SPX is analogous to TCP. SPX adds error correction and network information to each packet.

There are two versions of Novell's Netware. They are commonly referred to as Version 3.x and Version 4.x. The main difference is scalability. Netware 3.x supports 250 users. If there is more than one server, users have to log on to each server separately. Netware 4.x scales up to 1,000 users. In Netware 4.x users log onto the network rather than individual servers. You need only to log in once even if there are many servers.

Netware was designed in the days of PC ATs and 80286 processors. Netware doesn't require a lot of system resources. If a network has older DOS computers connected, it will probably use Novell networking. Novell is modernizing their network products to provide better support to TCP/IP and Internet networking.

Unlike TCP/IP and Netware, Appletalk combines all OSI layers, from the Network Layer all the way up to the Application Layer together into one protocol. This makes the network a seamless extension of the operating system. Appletalk is also self configuring.

Appletalk networks automatically configure each computer with an ID number. The administrative costs of an Appletalk network are practically zero. Appletalk is known as a peer-to-peer network. You don't need a central server to manage a network. Appletalk runs on a variety of network hardware including Ethernet, Token Ring and even ordinary telephone wires connected to the Macintosh's serial ports.

Unfortunately, Appletalk is hard to implement on non-MacOS computers. The tight integration between Appletalk and the MacOS makes use of high-layer protocols that work differently or are completely absent on other operating systems. In recent upgrades of the Macintosh's networking software, support was built in for other network protocols like TCP/IP.

Building or Upgrading

a Network

Before going to the trouble and expense of building a network, decide if you really need one. Many businesses build expensive networks and end up under using them. Do a cost/benefit analysis. Many small offices and workgroups get along quite well on sneakernet, handing disks and writing memos instead of using email. Shouting across a room is still a wonderful way to get attention.

Once you decide to build a network, choose your network hardware and software. Most people use whatever their computer vendor recommends. VARs and system integrators are good sources for advice, but make sure you know what you want before asking. The only thing you can be sure of is that you'll end up with an Ethernet network. Token Ring is commonly seen in shops built around IBM equipment. If you use an AS/400 computer as a server or mainframe you'll install a Token Ring network.

Token Ring is good if you're going to put your network under heavy-duty use. Ethernet is faster but it slows down when a LAN has too many users. When building an Ethernet, break the LAN into segments. Connect the segments with bridges. A bridge connects LANs together. The LANs can be different. You can have a bridge between an Ethernet and a Token Ring. Because bridges have little internal logic, both networks connected to the bridge need the same network protocol.

With most networks, you need a server. A server is a computer dedicated to file storage and network administration. A server monitors network traffic, stores information about users and gives access to expensive networked devices like RAID systems and high-speed printers. Sometimes servers are dedicated to one task. If there's a lot of printing on a workgroup LAN, it may have a print server -- a computer set up to spool files sent to a workgroup printer. Dedicated servers can provide Internet access or fax services.

A router is similar to a bridge, but more complex. A bridge connects two different networks together. The networks can use different hardware but must use the same network protocol. A router is smart. It can convert a data stream from one network protocol to another. A router lets you connect a TCP/IP network to a Novell network. Network servers are frequently used as routers. Common uses include providing dial-up access for remote users and Internet access for a network.

You don't have to use the same kind of computer for a server as the workstations. You can have a Sun (Palo Alto, CA 650-960-1300) running Solaris acting as a server for a Macintosh network. Or you can have a Mac running Linux, a free version of Unix, acting as a server for Windows PCs. If the server runs the same software as the workstations it eases administration. Windows NT machines are usually used as servers for networked Windows PCs. Macs tend to be used as servers for networked Macintoshes.

The exception is networks with a lot of heavy-duty use running mission critical applications. Unix servers are used because of their speed, reliability and ease of customization. Unix is the server of choice for most TCP/IP networks. Unix is complicated. For light and medium duty networks most people find Windows NT more than capable and easier to use.

The network protocol you choose depends on what devices are attached to your network and your legacy equipment. New computers can use any kind of networking software. Most operating systems have networking software built in. This makes it easy to get the network up and running after all the hardware is installed.

Appletalk is the easiest network to administer. It's fully automated and requires only minimal administration. PCs don't have Appletalk built in and don't blend well with Appletalk networks. Even if you have a Macintosh network, it may be easier in the long run to use TCP/IP.

Novell is the original PC network software. Novell networks are easy to maintain. There are more Novell LANs than any other in the world. There are network devices that only work on Novell networks. If you want to network older 286 and 386 computers you must run Novell software. There are disadvantages to Novell. It doesn't easily provide Internet access. If networked users want to access the Internet, TCP/IP has to be provided as an additional protocol. This complicates network administration.

TCP/IP is used on the Internet. If you're going to build a network based on a single protocol, TCP/IP is the one to use. Administration is a little complicated. Each computer on the network has to be manually assigned an address. You have to make sure there are no address conflicts. TCP/IP is universal. All computers use it. All modern computers have TCP/IP networking software built in to their operating systems. If you can connect a computer to the Internet, you can put it on a TCP/IP LAN.

Related Articles:

• OSI and Protocol Layers
• Four Reasons To Build A Network