Networking Basics
Networking is defined as connecting computers so they can communicate with each other as well as share resources and peripheral devices.
The driving reasons companies network computers include the desire to:
* Access, save, share, and backup data files in a central location.
* Share peripheral devices including: printers, scanners, faxes, modems, and photocopiers.
* Control access to sensitive information.
Networking technology consists of hardware and software components.
The primary hardware components are the NIC (network interface card) and the wiring that connects the population of the network to each other. The NOS, (Network Operating System), is the software component that enables communications over the network's hardware. At the most basic level, anything you can physically touch is hardware (i.e. a computer, a floppy disk, cables, printers, circuit boards, ect....). Software on the other hand, is a little more intangible. Software will always live on hardware but you never touch it directly. For example, the floppy disk, hard drive, or backup tape you might have is hardware; the program/data stored on the floppy disk, hard drive, or backup tape is software.
Hardware
The network interface card (NIC) is typically a circuit board installed in a computer. There are also models that can be attached externally like the XIRCOM pocket ethernet adapters. (XIRCOM is a reliable brand name, but there are of course other manufacturers of possibly equal quality). An external network card will either attach to a serial or parallel port of the computer.
Wiring, or network cabling, physically links the nodes or components of the network to each other.
Network cabling will normally follow one of two wiring topology schemes - a star network topology or a linear-bus topology. Each topology has its strengths; A star network has great advantages over a linear-bus network in terms of fault tolerance and modularity. Many small offices get a quick and dirty start in networking by installing a linear-bus topology network. This is largely due to the fact that the network wiring is can be laid across the floor, behind desks and cabinets, making installation cheaper and permission from landlords to run wiring unnecessary. As a network grows, it often changes to a star topology - as expanding the linear-bus wiring becomes problematic. The star topology is much more common than the linear-bus topology, especially in a new installation.
Software
A network operating system, (NOS), expands the resources of a computer by creating virtual connections to physically remote hardware and software. For example, a user may sit at their desk and be able to print to printer, photocopier, or fax that they have no direct physical connection to. Likewise, the same user may have access to accounting software that lives on a fileserver, (read fileserver as centrally shared computer), many feet or even miles away from their desk. A NOS accomplishes this by convincing the computer it has a direct connection to these remote resources. The computer is convinced by the creation of 'mappings' that establish a virtual connection to the remote device, mimicking a physical connection to the device. Potentially, a networked user will have access to every printer or peripheral device on a network and every megabyte of central storage space on the fileservers.
For example, a user on a Novell Netware network will in addition to their 'local' A: and C: drives have an F:, G:, H:, etc..... drive. These 'remote' drives (F:, G:, H:, etc...) will be virtual drive mappings to remote resources. Likewise their printer ports are not limited to local hardware and may be 'mapped' to network printing devices.
Network Architectures
This document will discuss three network architectures in exploring the basics of networking:
* Ethernet
* ARCnet
* Token Ring
These architectures are labeled as 'OPEN' since you can run a variety of network operating systems on them.
Before discussing the characteristics and differences between Ethernet, ARCnet, and Token Ring, it's a good idea to list the different types of wire used for networking.
10-Base-T/ CAT5/UTP Cable
10-Base-T networks will run on CAT5, UTP cable. The category rating of the Unshielded Twisted Pair (UTP) wiring is based on the number of 'twists' per inch the pairs of wires have inside the cable's jacket. Commercial/Industrial network installations will use 8-conductor (4-pairs) CAT5 cable. CAT5 is also available in a 4-conductor (2-pairs) flavor. 4-conductor CAT5 has everything that ethernet traffic needs. The advantage of running 8-conductor CAT5, when wired to the correct IEEE standard, is that the same cabling will support many different types of traffic.
CAT5 wiring will have certain characteristics including:
* Unshielded Twisted Pair
* 24-gage wire
* 8-conductor cable (available in 4-conductor)
* Plenum Level 3 (i.e. when it burns it will not emit noxious fumes)
* Rated for 100 mbps. traffic.
* Signal degradation occurs at about 300 feet.
* Uses RJ45 connectors. RJ45 is a type of modular connector, similar to phone cable ends but several sizes larger.
CAT5 typically comes in 1,000' spools and costs about 10 cents per foot. It comes in a variety of jacket colors and looks very much like a thick phone cable. UTP wire is more fragile than coax cable and costs about 20% less than coax. The manufacture AND the installation of CAT5 cable have certain guidelines/specifications that should be followed. Per specifications, CAT5 will withstand about 30 lbs. of 'pull' when the wire is being run; This means that the cable should move fairly freely when being pulled through ceilings and walls. Otherwise, the cable will stretch and distort - and no longer meet specification - when subjected to too much 'pull'. Specifications also instruct installers to cut back no more than 2 inches of the outer jacket when attaching the line to a 'punch-down' block, (instead of a RJ-45 connector). Punch-down blocks come in many sizes. They are made up of groves/slots the individual wires are 'punched' into and look very much like what you will find in telephone closets. Some installers will cut back much more than 2 inches of the cable's jacket to make it "pretty" in their wiring rack. When they do this, they are installing the wire out of spec and may experience unpredictable communication problems. This is happens for a simple reason, when the wires are all inside the 'jacket' and twisted around each other in pairs, they are affected in the same way at the same point by external forces. Our environment is saturated with electromagnetism. Computers 'read' signals transmitted over the wire 'pairs' by measuring the difference between the value of the signal on each member of the pair. If one member of the pair is subjected to influence that the other member is not, this will affect the reading and distort the signal. For this same reason, CAT5 should not be laid directly on fluorescent lighting or parallel to power cables.
10-Base-2/RG-58/Coax Cable
Coax cabling is typically found in a linear-bus topology network. It is also commonly used as a 'backbone' segment of complex star network topologies. Coax backbones will link 'sub-networks' of a star network over distances less than or equal to 600 feet. RG-58 coax cable, used for network cable, is very similar in appearance to cable TV wire.
Please note that cable TV wire, (RG-59), is built to a very different standard than the 10-Base-2 (RG-58) coax cable. The impedance, measured in Ohms, of RG-58 cable is 50 Ohms. Cable TV wire (RG-59) has an impedance of 75 Ohms. In basic terms, this makes it 50% more difficult for data to travel over the coaxial cable used for cable TV. Some devices are more sensitive to this problem and will not communicate at all if the wrong type of cabling was installed. The reason for mentioning this is that many customers, and even some contractors, make the mistake of installing the wrong type of coax. It makes no difference how much time and money was spent on installing the wrong coax, RG-59 is not network cable and should be replaced if discovered. With that said, IPX/SPX networks (Ex. Novell Netware) will run on RG-59 but they don't run well and sensitive devices (like network printers) won't work consistently.
Coax wiring has a very simple construction providing a two-wire circuit:
* Core: Usually solid copper, but can be made of braided wire. This is the first half of the two-wire circuit.
* Dielectric: Serves to insulate between the core and mesh and gives the cable both form and strength.
* Mesh: Surrounds the dielectric and serves as the second half of the circuit. The metal mesh makes physical contact with the BNC connector or 'end' of the cable.
* PVC Jacket: Surrounds the mesh. Provides strength and mechanical/moisture protection.
* Signal degradation occurs at about 600 feet.
* Uses BNC connectors. BNC connectors have circular ends connecting with a post-and-sliding-slot construction.
10-Base-5/Coax Cable
10-Base-5 Coax is an older, thicker, more expensive version of RG-58. It is no longer a widely used material and is mentioned here only to define the term.
Fiber Optic Cable
Fiber cable resembles 10-base-T cable in appearance. Its composition is very different. In simple terms, it has a glass core the light travels over. Fiber is more delicate than other cabling and is still extremely expensive to install. The major benefit of fiber is the great distances it can cover while still delivering a clear signal. Fiber optic cable can be used as a backbone between sites that are miles apart. Laying fiber cable between sites usually involves contracting the task out to the local telco (telephone company). This paper will not discuss the specifications of fiber cable since there are a variety of types to serve specific applications.
The two main types deal with the transmission mode:
* Single mode transmission: Used for clear transmission over long distances (telcos).
[Max distance 18.6 miles (up to 30 km)]
[Used to bring the circuit into the customer site (BDF - Building Distribution Frame).]
[Core is usually 8 microns.]
* Multi mode transmission: Used for broad bandwidth (data) transmissions.
[Max distance 1.2 miles (up to 2 km)]
[Used to connect between internal points on the network (IDF - Internal Distribution Frame).]
[Typical core sizes are: 50, 62.5, and 100 microns.]
The construction of fiber cable includes:
* Core: The transmission medium light travels over.
* Cladding: Denser glass applied directly over the fiber core.
(Maintains the core's transmission integrity)
* Buffer: Applied directly over fiber core and cladding to give strength.
* Kevlar: Surrounds the buffer for strength and mechanical/moisture protection.
* PVC Jacket: Surrounds the Kevlar for strength and mechanical/moisture protection.
It is now common practice to use twist-on ends for fiber cabling. This is a HUGE improvement over the older way of making ends on fiber cables. In the past, the process was very much like taking two human hairs, cutting the ends of exactly perpendicular to the central axis, polishing those ends to mirror perfection, and then gluing the two polished ends together in perfect alignment. Please note, even most contractors still buy pre-made fiber patch cables in lieu of making them.
Ethernet
Ethernet is by far and large the most common network architecture encountered today. Ethernet usually runs on 10-Base-T (UTP) wiring but in older installation might be found on 10-BASE-2 or 10-BASE-5 (Coax cable). Ethernet can be deployed on either a star or linear-bus wiring topology.
Ethernet: Linear-bus (coax) topology
Ethernet can be run on a linear-bus topology using coax (RG-58) cabling. This type of network will employ a point-to-point structure where every node will speak over the same cabling segment. Basically there will exist a single string of cable that will 'touch' each node of the network. (A 'Node' is defined as any fileserver, workstation, or peripheral device attached to the network). Each end of the 'string' will be 'terminated' to form an electrical circuit for communications to travel over. This topology is a simple two-wire circuit - the coax core and the wire mesh beneath the outer casing. A linear bus topology relies on the integrity of the entire length of its coax line. The major drawback to this structure is that if any point on the 'communications string' develops a fault, the entire network goes down. There are no communication 'Hubs' in this structure to use in isolating problems with specific segments of the network; the network is all one segment.
Note: Coax has a distance limitation of about 600 feet (before the signal degrades). Since a linear-bus topology treats the entire network as one segment, the distance between its two nodes furthest from each other cannot exceed 600 ft. The way around this limitation is to compensate for signal degradation by installing repeaters along the network. Repeaters will 'boost' the signal and send it off ready to travel another 600 ft., if necessary. This changes the scenario to one where the node must be less than 600 ft. from the nearest repeater instead of the furthest node.
Ethernet: Star/Complex Star (UTP) topology
A star topology involves 'hubs'. Hubs come in various forms and , depending on their complexity/abilities, may be called 'concentrators', 'switches', or 'ethernet switches'. The hub serves as a central connection point for the population of the network. Nodes, (fileservers, workstations, and shared network devices), will all 'plug-into' a hub to communicate with the other nodes on the network. Looking at the lines connecting the nodes to the hub from above you might see a star shape. That is , a picture similar to what you would get by drawing several lines through a single point to form a Christmas-card star.
Working with this star topology has many benefits in the areas of trouble shooting and fault tolerance. Instead of a single network segment touching each node, (linear-bus), each node has a separate cable segment that 'touches' the hub. This segment will take form, at least in part, as a 'patch cable'. Patch cables are lengths of network wire with connectors on both ends. Workstations will typically have a patch cable running between the computer and a wall outlet. Commercial/Industrial installations will also have 'patch panels' in the 'computer room'. Patch Panels represent the other end of the cable running from the back of the node's wall outlet. Yet another patch cable connects a node's 'port' on the patch panel to a hub. Because nodes have their own cable segment and speak to each other via a hub, any individual connection can be interrupted without affecting other nodes on the network. (The disconnected node will obviously be unable to reply to connected nodes but connected nodes can continue talking to each other). This is a significant advantage when trouble shooting network problems and 'isolating' traffic problems.
For example, network cards or connectors will sometimes become faulty. A faulty NIC or connector can cause normal data packets to become distorted and be broadcast as 'garbage'. In the case of a faulty NIC, it may continuously broadcast these garbage packets - flooding its own cabling segment (and probably those of its neighbors) to the point where normal traffic is unable to travel. This condition is commonly known as 'packet storm' or 'data storm'. Network administrators are often tipped off to this problem by finding that communication problems happen only at certain times of the day or only in certain departments. Many administrators have stayed late or come in early to diagnose problems and found nothing. They later determine that the problem only occurs when Jack/Jill is at work (and their computer is on). (Its also a good idea to take note of who is on vacation and who doesn't turn off their computer when they go home). The advantage for star networks in this case is that nodes can be connected/disconnected one by one to find the offending node. Connecting/disconnecting nodes on a hub is a simple as unplugging a phone cord. The normal scenario here would be to have two stations connected in the problem area, generate traffic between them, and add stations until the communication problem is duplicated.
Note: While IP nodes may let you plug and unplug them at will, IPX/SPX (Ex. Novell) networks need you to re-boot the node after disconnecting it. This is because IP is a packet-oriented technology and IPX is a broadcast technology. In IP, the packet is more or less self-contained and addressed to a specific destination. In broadcast technology, IPX, every node basically 'listens' to all traffic and only 'pays attention' to the traffic addressed to it. In ethernet, a 'collision detection' system ensures that if two nodes broadcast data at the same time and data 'collides', the data gets rebroadcast. IPX/SPX networks have several layers of 'drivers 'loaded to establish this 'listening' environment. When the physical connection is disrupted in IPX, the drivers need to be reloaded to re-establish the 'listening' environment.
Complex Star topologies
Complex Star topologies consist of multiple Star networks connected to each other. Complex Stars typically develop out of formerly isolated departmental networks. The consolidation of the individual star networks may be dictated or desired to provide one bigger/better central fileserver serving all departments in lieu of deploying smaller, less expensive, fileservers for each department. Consolidating networks also helps departments share information and physical resources (like printers). Individual stars of a complex star network are connected via a 'backbone'. Backbones are simply another cabling segment of the network. Backbones differ from other segments by the fact that they normally connect hub to hub instead of node to hub. Characteristically, backbones are often longer than other cabling segments since they might provided connectivity to stars that are hundreds or thousands of feet apart.
Crossover
Before describing the common types of backbones, we should discuss 'Crossover'. Most hubs will reserve their first port for backbones. This is significant because the first port will allow you to cancel the 'cross over' on that single port. Cross-over is necessary for effective communications. In ethernet, wires 1,2,3,6 are used for communications. Two of the wires or 'pins' are used to transmit data and two are used to receive. At some point, the transmissions of one node have to end up on the receiving pins of another node. This 'crossover' happens in the hub. Hypothetically the crossover should happen only an odd number of times. In practice, it only happens once. For this reason, you cannot simply connect one hub to another without turning the crossover 'off' on one of the connecting ports. A backbone simply combines two hubs into one large hub - typically over some significant distance.
Backbones
UTP: For connecting networks IN THE SAME BUILDING, running 10-Base-T is a nice, quick networking solution. However, the UTP cable should not be run through harsh environments, like extremely hot/cold warehouses or places where birds, squirrels, or pests might chew on or damage the cable. In short, UTP is very fragile. All an end user has to do to ruin a cable is run the wheels of a desk chair over it a few times. Signals will degrade after traveling over 300' of UTP wire and become unreliable. Repeaters can be installed compensate for the 300' distance limitation.
Note (repeaters): Although similar in appearance to hubs, repeaters serve a different function. Repeaters will take the fading signal from either side and 'boost' it to full strength, enabling it to travel potentially another 300', over CAT5/UTP. A repeater used in a coax backbone would allow the signal to travel another 600'. This difference in distance is a property of the cable more than of the repeater.
Coax
Coax backbones are quite common due to the greater lengths they can carry a signal before it degrades. In addition to having twice the range of UTP, they are less sensitive to their environments and fairly sturdy. Coax backbones can be buried, (preferably in metal conduit), or run in the open air between buildings, (ideally attached to a structure or rigid line). Coax lines can also employ repeaters to cover distances greater than 600' but its important to note that repeaters installed outdoors would have to live in a weatherproof workbox and be secured to a structure or tower. Its generally a bad idea to install a work box at ground level unless its immediately next to a building - vehicles tend to run over them otherwise. Many hubs will have a BNC connector on the back to support coax backbones. In this case the 'crossover' mentioned above is not an issue. If a hub does not have a 'native' BNC connector, a balen may be used to convert the RJ45 port to a BNC connector, (of course, one is needed on each end of the backbone). As implied just now, a balen will translate an RJ45 connector to a BNC connector.
Fiber-Optic Cable
Fiber will connect sites that are miles apart but fiber is expensive. The expenses include: hardware on both ends to connect to the fiber, paying the telco to lay the cable - if leaving the company grounds, and the fiber itself. Fiber is fragile (its essentially just glass) and should be professionally installed if going any great distance. The cost of fiber usually prohibits it use for short distances. This leaves a local network administrator choosing professional installation for most fiber. Its not uncommon however, to connect hub/switches in wiring closets with fiber when convenient. A ten foot, prefabricated fiber patch cable might cost about $35. Higher-end hubs will have slots that you can 'fill' with cards that support fiber cabling. The overall advantage of fiber is the high quality of the transmitted signal over great distances.
Wireless backbones
There are means of connecting buildings with line-of sight wireless backbones. Wireless backbones are basically a radio-frequency link between 'hubs'. This type of backbone is reliable - until you have bad weather or a bird takes a liking to your transceiver or a tree grows a branch in your line-of-sight. However, sometimes the customer just wants one because they know someone who has one or saw it in a magazine. Wireless backbones tend to need more technical support that standard backbones.
ARCnet
ARCnet is an old, old networking architecture. It is coax-based but employs a specific type of coax. Where it follows linear-bus wiring topologies similar to ethernet it has intricate rules regarding its cabling. Cable segments in an ARCnet network can vary form 100' to 20,000 feet - if you follow the rules. These rules, 'the Hierarchy of Hubs' involves both active and passive hubs. ARCnet is really much more trouble than it's worth considering that ARCnet is rated at about 2.5 mbps. vs. Ethernet at 10 or 100 mbps (megabits per second). In addition to the wiring rules, ARCnet uses 93 Ohm coax vs. the 50 Ohm coax of Ethernet. Currently there are few, if any, 'new' users of ARCnet. Those that do have it already are phasing it out.
Token Ring
Can you say I-B-M? Token ring was a product of Big Blue for large companies (with deep pockets) that needed complicated networks. Basically there was one 'token'. The token was passed from node to node. If you had the token you could talk to the other nodes - if someone else had the token, you just had to wait until the token got back to you. This is the exact opposite of a 'broadcast' technology (Ex. IPX/SPX). Token ring networks are built on the star topology. However, Token ring networks use SHIELDED-twisted-pair (STP) vs. UTP (Unshielded Twisted Pair). They also use hubs but not the same hubs found in Ethernet networks. Networks running Token ring will have MAU's (Multi-station Access Units) or , in plain terms, an IBM hub with and IBM price tag $$$. Token ring still has a presence in the industry, but you won't see many new token ring installations. This goes hand in hand with all the AS400s still in use because of the software they run. Many AS400 systems can today be replaced with powerful PCs for a fraction of the cost. In short, they were the best in their time, but technology got smaller, faster, and cheaper.
Differences between architectures
Bandwidth
The major difference of these architectures is in bandwidth. Please note that bandwidth is a measure of potential rate data can be transmitted over a network. There is always a bottleneck somewhere in a communication process. One of the two computers communicating may be slower than another, have inferior resources compared to another, or be performing more tasks than another.
Bandwidth vs. Throughput
Throughput is the actual speed data will transfer at from one point on the network to another. The actual speed, (throughput), of a network is often quite less than it's Bandwidth, (potential speed). Bandwidth is more often quoted than Throughput since it is easier to calculate and not subject to changes in many variables. The hardware used to communicate, heavy/light traffic, etc...can effect throughput.
Ethernet
Ethernet comes in two flavors of bandwidth 10 mbps and 100 mbps. Fortunately, (for 10 mbps networks that are upgrading), CAT5/UTP cable is already rated for 100 mbps. Ethernet running on Coax is limited to 10 mbps.
ARCnet
ARCnet is limited to 2.5 Mbps and no longer in wide use. There once was an effort to produce 'high-speed' ARCnet. That effort met little success and it is hard to find a mention of it in ARCnet reference materials.
Token Ring
Token Ring comes in both a 4 mbps and 16 mbps version. The 16 mbps version requires STP, shielded twisted pair, which could easily cost two or three times the price of UTP. The, much older, 4 mbps version would allow the use of UTP - but it was only 4 mbps. The major drawback to running 16 mbps Token Ring over 10 mbps ethernet was the token passing. Envision this, each NIC is responsible for passing the token on to the next after its predefined turn is up. If just one card malfunctions and fails to do this, the token could be lost and all the other cards will sit there still waiting for the token. For this reason, Token Ring networks typically have large support staffs.
Summary
Overall, running Ethernet in the 10 or 100 mbps flavor on a Star networking topology is the most flexible and reliable choice. However, in any information system, the software should drive the hardware decisions. If a company can not run with out software that only works on some old Unix system - they will need hardware and (at least part of ) a network that supports that Unix system. In this scenario, standard workstations can still be deployed using terminal emulation software to connect to the Unix system. The terminal emulation software 'pretends' to be a VT100 or 'dumb tube' station previously used to access the custom Unix software.
Common Issues to all Architectures
Network connections (NICs)
The selection of NIC will be determined by you architecture; Token Ring networks require Token Ring cards, Ethernet networks require Ethernet cards and so on... Ethernet cards are the most common and we will discuss those in this section.
Bits
NICs evolved in many stages. There were 8-bit, 16-bit, 32-bit and now 64-bit cards available. Basically, the more bits the card would support in communications, the better the potential communications between processor and LAN/WAN. Here 'better' is a mixture of both speed and quantity. It might often be the case that the data you transmit on a 32-bit card is no larger, at a given time, than you would transmit on a 16-bit card - but you have removed a potential bottleneck for times when traffic demands are higher.
Bus
Bus refers to the 'pipe' used to funnel data back and forth from the CPU of a node to the NIC. Common terms you might see are (listed roughly oldest to most recent):
* ISA: a.k.a. AT bus. (found in older 486s and lower speed machines)
* MCA: IBM's own - and very proprietary - flavor of bus. (Currently not in wide use). MCA cards are bus-mastering card. Bus mastering cards were a somewhat successful attempt at increasing the efficiency of communications between the card's slot on the motherboard and the processor. Basically, read 'bus-mastering' as 'proprietary hardware architecture'.
* EISA (Expanded Industry Standard Architecture): High speed interface still popular but usually found only in file servers, especially Novell fileservers, due to a limited selection of cards on the market made for the EISA type motherboard slot. EISA cards are also bus-mastering cards. EISA offers 32-bit throughput and auto-reconfiguration. Both EISA and VESA incorporate principles of FLEX. FLEX was originally developed by Compaq Computer Corp. in an attempt to increase the efficiency of communications between peripheral components and the processor.
* VESA Local Bus: This short-lived technology was essentially a direct line between peripheral slot on the motherboard and the processor (very similar to Macintosh bus technology). (Macintosh developed PDS, (processor direct slot), and NuBus both of which involve a direct connection between the processor and peripheral slot.)
* PCI: Currently enjoying wide acceptance as the bus of choice in the PC industry. PCI slots are found in late model 486s and all Pentium systems.
Memory access methods
There was a time, pre-windows'95, when memory management played a greater role in installing network interface cards. In fact, it was only really in versions 6 and higher of MS Dos that we started to see memory management integrated into a PCs operating system. The NIC needs to communicate to the processor through a 'piece' of memory. Typical methods of this are:
* Shared memory - uses up memory that could be put to better use and often causes conflict with other hardware trying to talk to the processor.
* Direct Memory Access - a little better than the 'shared' method, but it uses up a DMA channel that other devices might need to talk to the processor.
* Programmed I/O - you might also call this a memory-mapping card - it has its own little piece of memory to add to the computer and gives the best overall performance of the lot.
The Network Interface Card
The NIC will come in three general types: internal, external, and PCMCIA:
* Internal: By far and large the most common. 8-bit, 16-bit, 32-bit, 64-bit, 10 mbps, 100 mbps, etc... These cards are simply another circuit board that lives in a 'peripheral slot' on a PCs motherboard. Peripheral slots may be filled by many items including: modems, video cards, and controller cards.
* External: External interface cards were big before PCMCIA slots hit the market. They are about the size of a package of two Swiss Rolls or a PC's mouse. External cards are very nice for technicians since you can install them temporarily without opening a computer case. It also saves time diagnosing hardware problems with internal network interface cards. If the network connection works with the external card but not the internal card, it really narrows down your search for the problem.
* PCMCIA: Personal Computer Modular Component Interface Adapter? Or how about 'People Can't Memorize Computer Industry Acronyms'. Maybe we should all call them credit card adapters since the shape and size are just about the same. PCMCIA slots are externally accessible peripheral slots. Available PCMCIA devices include: NICs, modems, controllers, hard drives, security access cards, etc... The advantage is that these devices can be added (and removed) without opening the computer's case and many times without even shutting it off. However, installing certain PCMCIA cards involve a complex process of driver loading.
Network operating systems
Client/Server
The client server model uses a central computer, (fileserver, main frame, master node), that 'serves' files to the clients/nodes. Fileservers are computers whose only function in life is to server the clients/nodes. No one actually sits down at the fileserver to do his or her daily work. The application programs loaded on the fileserver are accessed from one of the clients and not directly at the fileserver. In the old days, the clients were simply 'dumb tubes'. They had a monitor and a keyboard. This type of system is still common in banks and hotels were storage of data locally at the client is undesired or unnecessary. The trend has been to use 'smart' clients in most new client/server networks. Smart clients will have their own local storage of data and programs and offer far more flexibility. The advantages to the Client/Server structure are that security, data backups, and device sharing can be administered centrally.
Good examples of client server networks include: Novell Netware and Unix. There are many variations of Unix out there, some of which are proprietary and only run on the hardware supplied by the vendor.
A network with smart clients has many operational advantages due to the fact that storing applications and data locally (when appropriate) will take large burdens off the central/shared computer. For example, if everyone on the network uses the same word processing or spreadsheet software, it is operationally better to store the program files on the workstation/client. At this point, it is necessary to distinguish between program files and data files. Program files are what you buy from the vendor. These are the files that exist after the initial installation of the program. Data files are the files you create with the program. Databases or accounting programs are significantly different from other applications because the data created with them is integrated into the application. To clarify, a user can write a letter and save that document on either the workstation or the fileserver. The same user might even copy it to a floppy disk and walk away with it. A user cannot copy an 'account', (i.e. vendor or customer file), from the accounting software on the network to his hard drive or floppy disk. Loading application software locally on the clients avoids huge amounts of network traffic every morning when, otherwise, each user would load need to load it from the fileserver. In contrast to this, accounting programs would not want to store data on local hard drives. The difference between these scenarios stems from the nature of the data. Word-processing and spreadsheet programs do not change from day to day. Accounting programs do change everyday. Efficiently structure networks typically store 'static' programs locally on the smart clients and the data, (documents, spreadsheets, etc...), generated by these programs on the fileserver. Likewise 'dynamic' applications normally reside on the fileserver since maintaining multiple copies of databases becomes unfeasible very quickly. The most important fact to keep in mind is that if you want the data to be 'backed up" it should be on the fileserver. The rule of thumb is to back up you data as often as it would be inconvenient to lose it. If you only enter new data into the system once a week, then backing up once a week is sufficient. Most networks make backup copies of data at least once a day.
Peer-to-Peer
The second type of network is peer-to-peer. The significant characteristic of a peer to peer network is that there is no central computer. Each node may have programs or devices it offers to share with other nodes but the configuration of security, backups, and device sharing is dependent on the individual node(s).
Peer-to-peer networks often run on a linear-bus wiring topology. Many small companies will get their feet wet in networking with a peer-to-peer network. They find this type of network attractive because they avoid the additional expanse of a fileserver and coax cabling can just be dropped behind everyone's furniture to make the job quick and easy. There are peer-to-peer networks that run on CAT5 using a star networking topology. Star networks necessitate a hub - another expense customers avoid by going with linear-bus. Peer-to-peer networks typically have huge security holes since users and change their own settings. They also characteristically need more technical support for the same reason.
Good examples of peer-to-peer networks include: LANtastic, Netware Lite, and MS Win'95 networking.
Hybrid
Microsoft builds peer-to-peer networking into their operating systems from the ground up. NT networks are a perfect example of a hybrid network for this reason. The two systems essentially run on top of each other. Users can share files on their workstations in addition to the resources available to them on the network. The focus of the network however, remains on the fileserver and users are normally discouraged from using the peer-to-peer functionality available to them.
Device Sharing via Network Operating System
A key driving force in companies' decisions to network their computers at all is the sharing of devices. The two most beneficial devices to share centrally are hard drive space and printers. The central sharing of hard drives enables the company to invest in one large, high quality, fast hard drive system. It also permits the company to store all important data in a central point to make backing up (and restoring lost data) much easier. Central data storage also permits the use of central application programs like integrated accounting software and databases. Sharing printers via a network is usually the first benefit of networking realized by end users. For example, instead of buying a dozen inkjet printers to deploy at a dozen workstations, the company can buy one or two high performance, high quality laser printers and make them available to everyone.
Expected Networking Costs
Hardware:
The most significant cost of networking will be a fileserver. It is often hard to convince customers that this single device should cost so much money. A common customer response is "Why should I spend so much money on a computer that no one uses?". The key in serving the customer best in this scenario is to explain that EVERYBODY uses the fileserver and money 'invested' in this device benefits the entire population of the network.
There are actually formal calculations for how much disk space and memory a file server will require but they are dependent on a number of factors and will not be discussed here. As a very general rule of thumb you can assume 32 MB of Ram for the fileserver operating system (Novell or NT) and an additional 4 MB of Ram for every Gigabyte of hard drive space on the fileserver. Add to that to that another 4 MB for every major application running on the fileserver all the time. Take that total and round up to the closet number/size combination of memory chips your server will hold. Most motherboards will hold 4 memory chips. Memory chips commonly come in 8, 16, 32, and 64 MB sizes and are normally added two at a time.
A reasonable fileserver, (hardware and installed operating system), will cost between five and ten thousand dollars.
Network interface cards can run anywhere from $25 to $125 each. They are cheapest if purchased in bulk and are usually available in 10-packs.
Wiring will be cheapest is purchased by the spool. CAT5/UTP wire comes in 1,000' spools and runs about 10 cents per. foot. Coax will come in 500' spools (or larger) and costs about 13 cents per foot. Shielded wire comes in several types and is easily two to three times as expensive as the UTP wire.
Software
After installing a network, a company will typically want to upgrade existing software packages from a stand-alone version to a network version. Upgrading the existing software licenses to buy the network version is much less expensive than purchasing the network version outright. Even if the users have a mixture of software this is still possible. Most vendors will allow 'competitive upgrades'. In a competitive upgrade, you essentially trade in your license on vendor A's product for a license on vendor B's product. This is in vendor B's best interest since, in theory, you agree to stop using vendor A's product and destroy your copy of it. That in turn leave one less user in the market with vendor A's product and one less copy of vendor A's product to generate the sale of the upgraded version of vendor A's product. The licensing game should be played honestly and carefully. Many companies who play loose with licensing have been forced to pay outrageous penalties, fines, and damages to software vendors. All it takes is one disgruntled employee to report the firm to the software police.
Training
There are several approaches to the training process:
* Buy just the hardware and software and make some poor soul on staff learn to install it and teach themselves (and everybody else) how to use it in addition to handling their existing workload. This typically involves spending long hours waiting on hold for free tech support.
* Hire a new person and have them follow the same steps listed in the first example. The difference here is that the will not have the pre-existing workload of the person in the first example.
* Hire someone knowledgeable in networking and make the installation of the network and user training their priority. After the network is up and running move them over to some accounting functions to keep them busy until the network breaks. Historically, the MIS functions grow out of the accounting department since they tend to be the initial users of computers and networks in an organization.
* The mid-range solution is to take an existing person knowledgeable in how the company already operates and send them to some network training. This is the best of both worlds since this person can in theory help integrate the network into the existing operation with minimal disruption. This also saves the significant cost of hiring consultants or solution providers.
* Hire a trainer to come in and teach some or all of the staff. If work duties don't allow everyone to attend at once, those who do attend might be assigned to pass the knowledge on to those that didn't. Otherwise, you can keep the trainer for a day and have separate morning and afternoon sessions. It is generally a bad idea to hold this training after hours since workers are thinking about where they would rather be.
Training can be included in a 'package' of hardware and software purchased from a vendor. It is best to specify in writing with the vendor exactly what type of training and how many hours of each type will be provided. If you don't have it in writing, you will usually end up not getting as much training as you expected. This is often because the customer and consultant have different perceptions of how much training is needed.
Large authorized resellers, (VAR's - value-added-resellers), will often have training centers and scheduled classes. Training at one of these centers typically runs $100-$300 per. full day (depending on the course). The network administration courses are normally 3-4 days long and priced in the $300/day range.
Consultants
Like any other profession, there are good consultants and bad consultants. Please keep in mind that there are a lot of consultants running around with 'certifications' for products or operating systems who have never done a customer installation. There is also a difference between 20 years of experience and 1 year of experience lived over 20 times. If you can find a consultant who has both experience and certification in the product he recommends, this is wonderful. It is not uncommon to find a consultant that has installed and maintained 20, 30, or 40 networks but never bothered to sit for the certification exams. In short, get references!
Consultants range in price depending on the area of the country and the number of products they are authorized to resell. The more products a company represents as a VAR, the greater their annual expenses and the more they must charge. Consultants are a bargain at any price - just ask one!
Typical hourly rates for consultants range from $65/hr. to $125/hr.
The Cardinal Rules of Consulting
1.) The consultant is a bargain at any price.
2.) If the customer is not complaining, you are not charging enough.
3.) If you can't fix it, it must be a hardware problem.
4.) If it's a hardware problem, the customer must have broken something.
5.) They key to any profitable consulting relationship is being to tell the customer that you've always been right and they've always been wrong (and getting them to thank you for it).
(
geocities.com/siliconvalley/peaks) (
geocities.com/siliconvalley)