|Volume 20, Number 3||1995|
Two years ago, the College of Veterinary Medicine at Kansas State University (CVM-KSU) adopted seven goals for computing: 1) to establish a college-wide network; 2) to provide faculty, students, and staff with either their own microcomputer or easy access to microcomputers attached to the network; 3) to automate the keeping, retrieval, and analysis of hospital records over the network; 4) to instruct students in the use of computers particularly in veterinary practice management; 5) to provide faculty with the hardware and software necessary to develop interactive multimedia lessons; 6) to provide students with 24-hour network access to the interactive multimedia lessons; and 7) to establish a means for clients of the College to access information over the network.
This paper describes the strategies used to accomplish these network goals so that others might profit from our experiences, especially college administrators and other leaders and network planners and managers. First, the benefits from networking our College are presented, then the essential elements of a college-wide network, their function, and their cost are presented (1). The philosophy that guided our selection of components for the CVM-KSU network is discussed. Other topics discussed include cabling systems; how computers and printers are attached to network cable; software used to operate and manage networks; services provided to users of a network; network performance; ease of use; network user training; and how network support staff might be organized.
As network technology has developed, a language has developed along with it that must become familiar to college administrators who must make decisions about the extent of network support to be provided and to network implementors who must make decisions about the hardware and software required to meet the desired level of support. Therefore, the terminology of networking is used in this text. If the meaning of a term cannot be derived from the context in which it is used, the reader can look up the term in the glossary that is provided at the end of the text. The glossary also is provided as a stand-alone quick reference guide so that the reader does not have to search through the text to find a definition.
We have found that network technology has been developed sufficiently that industry standards have been established for network hardware and software, making an investment in client/server-based computing less expensive and less risky than it was previously. Central mainframe computers are being replaced by highly integrated client/server-based networks. A server is a computer that serves other computers and devices on the network. Much like a server in a restaurant who serves a meal to a client, a server on a network serves a file to a client. A client is another device on the network, such as a user's computer, a printer, a modem, or another server. Because computer files (both program and data files) are being served, servers are sometimes called file servers. If the server serves only a particular type of file, it may be called, for example, a print server or mail server. A file server only delivers files to the memory (RAM) of the user's computer. It is the user's computer that actually runs the program delivered or manipulates the delivered data. The delivery time of a file server is not noticeably different from the time it takes to copy a file into memory from the disk of a stand-alone computer. Servers are less costly to purchase than a mainframe computer, are less expensive to maintain, and are generally more user-friendly. An added advantage of a distributed system in which software and data are kept on several servers instead of one mainframe is that if one server should fail, another can replace its functions, sometimes even automatically.
Impediments to Building a College-wide Network
Veterinary colleges often experience one or more of the following real or imaginary impediments to developing a college-wide network: 1) the large expense associated with implementing the network; 2) an existing investment or debt in computer equipment; 3) a bad previous experience; 4) lack of personnel with computer networking expertise; 5) insufficient faculty or administrative pressure to make the establishment of a college network a high priority; 6) uncertainty about how to implement and administer a network; and 7) lack of training for users of a network. Nay-sayers abound and claim that a college computer network will create bureaucracy; inhibit personal freedom; and be costly, slow, dangerous, unreliable, inflexible, restrictive, and anything but confidential.
Benefits of a College-wide Network
We have found many benefits to having a college-wide network. For instance, our users are freed from software purchasing and maintenance; have more disk space; have access to software and hardware otherwise unobtainable; have their files automatically backed up daily onto magnetic tape; have easy file recovery; and are able to send, without charge, files and electronic mail to others on the network and to people on networks who are thousands of miles away through transparent access to Internet (an international network). We have found that sharing devices and software reduces cost. Rather than placing an expensive computer on each person's desk with all the peripherals and software that each person needs to operate efficiently, a less expensive (even diskless) computer is placed at each workstation with fewer or no peripherals. Computer personnel no longer have to update software or backup data on each computer in the College, only on the file servers. Databases do not have to be copied onto several computers, which creates concurrency problems, because the network allows people working on several computers to share common files. The network allows users to connect to computer resources at work from home. The network is easy to use because it gives users transparent access to the network regardless of the cabling used to connect the user of the users' type of computer. Often users are unaware that they are accessing files on the server and not on their own computer's disk drive. The network enables groups of previously disparate computers and printers to become a single integrated system. The users' computer resources now include everything on the network and not just what is on the desktop.
We have used the network bulletin board to establish Electronic Learning Communities (ECL). For instance, faculty of the Food Animal Learning Community post questions on the network to be discussed on-line such as "What is the responsibility of food animal practitioners to the public on issues such as food safety, environmental preservation, and animal welfare?" Thus, the network has provided more than a means for connecting computers, it also has provided ways for people to connect to each other: students with students, students with faculty, faculty with clients.
The network has improved communication among personnel and has reduced the time and cost of photocopying and distributing memos. It has helped break down geographical and organizational boundaries. The network has improved efficiency of communication between the Dean's office, business office, and hospital, which are housed in different locations but must interact frequently and share files. The network has sped up preparation of research grant proposals by allowing collaborators to work on the same wordprocessing document easily and has facilitated management of research grant accounts.
Messages between departments can be sent faster (and sometimes more reliably) by electronic mail than by campus mail. Messages that may have been too long or too confidential to leave with a department secretary now can be sent electronically directly to the person addressed.
Users will complain if programs run slower on the network than they do on a stand-alone machine. We have had no complaints by users about network performance and no obvious degradation in speed even when students in a 35-station computer laboratory were accessing large statistical applications simultaneously.
The network was easy to install (it took 2 people from the contractor 3 days to install the entire network infrastructure and an additional 2 weeks for our personnel to set up the file servers so that the network was usable). The network has been easy to maintain and expand. The network has provided so many benefits that many faculty consider it to be as necessary to their productivity as their telephone or the library. These faculty and the administration now believe that a network and a microcomputer should be provided to all faculty as a matter of course. The most important advantage of the network is that it has provided everyone with equal access to computing resources-an important equality of opportunity.
Design. We did not conduct an extensive, in-depth, user needs analysis or a cost/benefit analysis beforehand; instead, we assumed that a college-wide network would be helpful and proceeded directly to the design stage. We paid no consultant fees to design our network (we could not afford their fees and did not know if we could trust their advice). Instead, we spent time educating ourselves. We made a list of potential sources of free information: the campus computing center; the campus telecommunications center; computer manufacturers, such as IBM (2), Apple (3), DEC (4), and SUN (5); telecommunications companies, such as Southwestern Bell Telephone (6); telecommunication equipment manufacturers, such as SynOptics (7); and value-added retailers, such as Kansas Communications. We "picked the brains" of salespeople and technicians from these potential vendors and others until we knew enough to write the specifications to request bids on the network components and installation. We explored phone closets, elevator shafts, ventilation ducts, and crawl spaces above ceilings until we were able to draw a diagram of the proposed location of the network components and the means by which they would be connected. We learned a new vocabulary of words such as connectivity and topology; new meanings to familiar words such as runts and giants; nonintuitive acronyms such as TCP/IP and I0BASE-T; and encrypted sales jargon such as "standards-based multivendor interoperability."
Installation. Although our campus has a central computer center, we did not rely on them to install and manage our network, because they have other clients and responsibilities, and we would have little or no control locally. We felt that the campus computer center might respond slowly to veterinary users and, other than electronic mail, might provide few services with utility for veterinary medicine. Instead, we relied on one member of the faculty, three computer personnel already employed by the College, a private contractor, and a supportive college administration to build and maintain our college-wide network.
We provided training at a cost of approximately $5000 for one of our computer support staff to become a Novell Certified Technician, because no one on the campus was certified. Although this certification was not required (our network was installed before his certification), the training has helped ensure efficient network performance and growth. Our technician has access to people otherwise inaccessible with expert knowledge about Novell. We use our technician for maintaining the network and do not pay for a maintenance contract with someone outside the college.
The network is composed of computers, plotters, printers, and other devices connected by electronic cables and other electronic signal carriers; the services provided to users by these devices; and the software used to manage the communication among these devices. Computer networks allow people to share resources such as printers, storage units (e.g., disk and tape drives), data, and computer programs and to send messages to each other. Our selection of network hardware and software was based on a philosophy of providing users with personal freedom. We learned that we could not dictate what hardware and software faculty will use. We further recognized that the configuration of offices and laboratories changes over time. Typically, in large firms, 30 percent or more of all workers change locations each year (8). With network topologies and technologies evolving so rapidly, we designed the network for maximum flexibility. We wanted to attach any kind of computer (e.g., Apple, IBM, or Sun) to the network simultaneously. We wanted to use any type of cabling system; e.g., unshielded twisted-pair (UTP), shielded twisted-pair (STP), thin or thick coaxial cable, fiber-optic, or Attachment Unit Interface (AUI) transceiver cabling simultaneously. Finally, we wanted to use any type of network software; e.g., Novell NetWare (9), Banyan's VINES (10), IBM's Token Ring 11(11), 3Com (12), Microsoft's LAN Manager (13), Artisoft's Lantastic (14),or Apples AppleTalk (15)] simultaneously.
We wanted to support simultaneous network communication at speeds of 4 Megabits per second (Mb/s), 10 Mb/s, or 16 Mb/s and support high-speed fiber distributed data interface (FDDI) networks in the future (100 Mb/s). We wanted a network that would be cost effective to change and expand as user needs grow and as new techniques become available.
Network Cable. The type of network cable used depends on the desired speed and quality of transmission, cost, ease of installation and expansion, and the number and distribution of devices that are to be connected. We used a mixture of cabling types: fiber optic, unshielded twisted-pair, thin coaxial cable, and shielded twisted-pair. Fiber optic cable provides the fastest transmission rate, but is expensive. New technologies have increased the transmission rate through twisted-pair from 4 Mb/s to 10 Mb/s to 100 Mb/s, and the rate eventually should reach 640 Mb/s. Our system presently runs at 10 Mb/s primarily over unshielded twisted-pair wire. This is much easier to handle than coaxial cable because it is light, narrow, and flexible but not flimsy. Attaching connectors to twisted-pair wires also is easier and less expensive than attaching connectors to coaxial cable.
Unlike the telephone wiring in most homes which is a continuous line connecting each telephone jack, most office buildings have phone closets from which individual lines run to each telephone site. This allows testing and trouble-shooting of circuits at a central point. A central connection point, or hub, simplifies network access, moves, changes, and expansion and quickens identification and resolution of network problems. Unlike a daisy-chain installation, where a bad connection to one computer disables all "downstream" connections to the network, each port on a hub is connected point-to-point with a network device, such as a computer or printer. Any problem with the connection affects one device and one segment only, and the entire network does not go down if a segment fails.
Our network (Figure 1) consists of a fiber optic backbone connecting the 3 buildings of the CVM-KSU. The fiber optic cable connects 4 area concentrators ("intelligent" wiring hubs with information processing capability) which are housed in the phone closet for each area. These 4 phone closets serve all the telephone lines for the CVM-KSU; therefore, to attach a room to the network, all that needed to be done was to connect the unshielded twisted-pair phone line in the telephone closet to the concentrator housed there. Patch panels (Figure 1) were installed to enable connections between devices on the network and the concentrator to be changed easily simply by moving the patch cables (Figure 1). Punch-down blocks (Figure 1) were used to connect a port on the patch panel to an extra pair of strands of the telephone cable running to an individual phone jack. Because every phone line is now also a computer line, moving network users is as easy as moving their computers. No new cable must be pulled, so there is no delay. Because the wires and cables radiate out from a concentrator, the cable is said to be configured in a star topology. A topology is the physical layout of a network, including the cables and devices. Other topologies are a ring and daisy-chain. Two issues must be considered when designing a network: 1) the length of cable runs (signals lose strength and pick up static if the cable length is too long) and 2) the number of times the signal must be repeated (amplified and passed along the network by a hardware device called a repeater). By using a star-configuration that distributed the UTP wire from a hub, we never exceeded the maximum recommended distance, 110 meters (365 feet), for reliable communications over a segment of UTP wire. Although signals can be repeated 3 times without degradation in strength and quality, no signal had to be repeated more than twice with our star configuration.
We saved the cost of installing new wiring for much of the network because we made use of existing telephone cable. To be able to use existing telephone cable for a computer network, the cable must include extra pairs of wires in good condition and be configured as a star (e.g., separate lengths of cable run to each phone jack from a central location such as a wiring closet). Telephone cable usually has extra pairs of wire so another wire can be used if one breaks or other telecommunication devices can be installed at one location. Stringing new cable is costly, especially if there is an asbestos hazard in the building, if cabling must pass through plenum ceilings, or if cabling must fit into narrow spaces. New cables can be unsightly if they must dangle from ceilings or if conduit must be attached to wall surfaces to house them. Even when existing wire saves the initial expense of installing new wire, a star-configured network can be more expensive to install than a daisy-chain network of the same size, because the former requires hub equipment. For larger networks, though, the additional cost of a star-configured installation is offset by 1) the lower cost of adding users; 2) the ease with which changes to the network can be made; 3) the lower cost of network troubleshooting and repair (network managers do not have to walk from office to office to track down a problem); and 4) the reduced downtime because a separate wire segment is used for each node (e.g., a computer or printer), so a problem in any one node affects one segment only, and the remaining segments are available for continued operation. In our computer laboratories, where many computers are next to one another, we used thin coaxial cable (thin-net) to connect them to the network because it was more convenient and economical to run a single cable with individual computers attached along its length than to run individual lines to each computer from a concentrator or file server. These lengths of thin-net (Figure 1) were connected to one of the area concentrators, so users of the computer laboratories could access the entire network.
A fiber optic cable that had been strung from the campus computer center to the CVM-KSU was attached to our fiber optic backbone with a router (16) (Figure 1). The router is a hardware device that provides individuals on the College network with access to other campus networks and the campus mainframe computer, but prevents local network communications from traveling throughout the campus. It is called a router because, like a mailman who delivers mail to addresses on a particular route, a router delivers data to addresses on a particular network. The router stores in its memory the addressing information for each network attached to it. Each node on a network has an address.
The cable that had been strung to the college from campus was 12-strand fiber optic cable, but because we did not require this capacity at the time nor expect a need for this capacity in the future, we used less expensive 6-strand fiber optic cable. A fiber optic interface (FOI) is a wall-mounted metal box that provides a place to secure the ends of fiber optic cable in our wiring closets. The FOI also houses any slack (extra length) in the cable and provides strain relief (i.e., the FOI helps support the load of the cable and allows the cable to expand and contract as its temperature changes). We used an AT&T FOI to connect strands of fiber optic cable to: 1) a concentrator; 2) other strands of fiber optic cable when the cable had to be extended from one wiring closet to the next; or 3) to a terminator which is a required connection at the end of the fiber optic backbone. Installation of the concentrators and fiber optic cable was subcontracted to a private company 17(17). We used a transceiver (hardware that has an input and output cable port to receive and send network signals) whenever two types of cables (e.g., thin-net and twisted-pair) needed to be connected together.
Network Concentrators. In the center building of the CVM we used a SynOptics 3000 concentrator; Cabletron 18(18), 3Com 19(19), and Fibermux 20(20) are other brands. The SynOptics 3000 has 12 slots to hold plug-in electronic boards (cards) called modules. The modules have ports for connecting the concentrator to individual runs of cable. Modules can be equipped with ports for connecting unshielded twisted-pair wire, shielded twisted-pair wire, fiber optic cable or thin or thick coaxial cable. Thus, the concentrator can be adapted for any combination of cabling needs. We equipped our Model 3000 concentrator initially with 3 Model 3308 modules to provide 36 UTP ports, a Model 3301 module to provide 8 thin coaxial cable ports, and a Model 3304-ST module to provide 6 fiber optic ports. An additional slot was used for a 3313-02-A network management module. The network module continually collects activity, configuration, and diagnostic information about the concentrator, modules, and ports and forwards the information to the network manager's computer where it can be displayed graphically. An external modem can be connected to the network management module so that the information collected can be accessed remotely.
In the wiring closets of the other two buildings, we used Synoptics Model 2310 concentrators equipped with 36 UTP ports to connect to the telephone wires centered there. The 2310's were used because we did not need to mix cable types in these wiring closets, and they were less expensive than using a Model 3000 concentrator. When we had to increase the connections for these areas because more people wanted to be connected to the network, we used a cross-connect cable to link several Model 2310 concentrators together in the wiring closet. Between the initial installation and this writing, we have added 4 model 2310 concentrators (144 users) to our network.
An important feature of network concentrators is that all configuration parameters and status information are saved in batter-supported RAM, so all critical data are saved in the event of a power failure. LED indicators on the front of the concentrators and on individual ports report status information. The built-in Attachment Unit Interface (AUI) interconnect port was used with a Model 504-ST external transceiver to connect the concentrators to the fiber optic backbone. The transceivers also were equipped with status LEDs. The concentrators were wall-mounted in industry-standard aluminum 19-inch equipment racks which have space for adding additional patch panels of 2310 concentrators. The power supply and fans are accessible from the front of the concentrator for easy replacement. The concentrators are protected from power surges, short circuits, and overheating.
Network Cards. We installed a network card into an expansion slot of each computer on the network to connect the computer to the network cable. This is more convenient, less expensive, and more aesthetic than using an external transceiver. The type of card depended on the type of computer to be connected and the cable to be used. When purchasing a network card, the purchaser must be sure that the card will fit in the computer's expansion slot, be equipped with the correct cable connector, and be compatible with the software and hardware on the network. The last criterion is the most difficult to determine and may require trial and error evaluation of the card. We use NE2000 16 bit AT Bus network cards to connect DOS-based computers, Assanti network cards to connect Apple MacIntosh computers, and Cabletron network cards to connect IBM Micro ChannelÆ computers to the network. At the time of purchasing a network card, we specified the type of connector required (e.g., UTP, thin-net, IBM token ring). Some manufacturers provide several types of connectors on one network card. We selected these network cards after evaluating cards from several manufacturers. We found that cards from some manufacturers did not allow us to switch between our Novell network and our campus network without first shutting off the user's computer. The cards of some manufacturers failed to work at all with the software we use to connect to our campus network. Therefore, we recommend that purchasers evaluate a manufacturer's card before purchase and, to prevent possible incompatibilities, try to buy all cards for a given computer type from one manufacturer.
Network Operating Software. A network operating system (e.g., Novell, LAN Manager), not a workstation operation system (e.g., DOS, OS2, UNIX, Apple System 7), is the software used for managing the sharing of servers (e.g., file, print, mail) and applications (e.g., wordprocessing, spreadsheets, database managers). The network operating system is loaded on the file server. Novell's Netware, Version 3.1, was used primarily as our network software, but some of our laboratories have used LAN Manager, IBM Token Ring, and Lantastic. Our Novell license, which cost approximately $4000 including manuals, restricts the number of simultaneous, not total users, to 250. We connect to the campus mainframe by using Clarkson University's (21) public domain device drivers which use the TCP/IP protocol. A device driver is a disk file that contains the information needed by a program to operate a device. A protocol is a collection of rules for sending information over a network, such as its format, timing, error control.
File Servers. It was not enough simply to provide users with the ability to attach to the network, we had to provide them with reasons to attach to the network. Therefore, we attached 6 file servers (22) to the network that were loaded with useful and easy-to-use application software. Because many users may be requesting files at once, a file server should be fast at sending files. The server should have a fast central processor (>33 MHz); fast memory (<70 ns); and fast (<14 ms average access time), large capacity (>1 gigabyte), disk drives. Each of the 6 file servers we purchased was equipped with a 33 MHz Intel 80486 processor, 16 MB 70 ns RAM, 2 Fugitsu 680 MB SCSI drives with an average access time of 14 ms, a 1.44 MB 3.5 inch floppy drive and a 1.21 MB 5.25 inch diskette drive to copy software onto the servers, 1 parallel port and 4 serial ports to attach network printers, a monochrome monitor (a server does not require an expensive color monitor), and a keyboard. The cost of these file servers was approximately $3400 each.
Server Software. The application software copied onto the file servers included electronic mail (23), a bulletin board (24), spreadsheets (25), database managers (26), statistical packages (27-31), graphics (32), and programs for practice management and bibliography (33) plus numerous applications specific to veterinary medicine. The amount budgeted for purchasing network versions of this software was $5000. By placing this software on file servers, everyone has been allowed access to the latest version of the software, not a select few. Also software maintenance has been easy because the software has had to be installed only on one machine.
An issue to be considered is who determines what software should be loaded onto the file server and when it should be loaded. Users make requests for software directly to the Computer Services Group that manages the network and the group solicits suggestions from users periodically by electronic mail. When users purchase software themselves with university or extramural funds, we encourage and often subsidize the purchase of a network version so that everyone in the College can benefit. The Computer Services Group has been given the responsibility to determine what software is available that will help faculty, students, and staff do their work more efficiently. This is not an easy task, because computing technology changes rapidly and users' needs also change, so information on what software is best is always incomplete and inadequate. Criteria we use to help make software decisions include 1) whether the software will work on a network; 2) whether the software is compatible with the computers and printers we have; 3) the cost of the software (of course, because we teach the use of software to others who are likely to purchase the software themselves, we expect the vendor to provide the College with the software free or at a greatly reduced cost); and 4) how widely the software will be used (although we may purchase software that benefits only a single individual if that individual's benefit is great and if the software does not usurp other users' benefits).
We expect the vendor to provide us with an evaluation copy of software we are considering for purchase. We may not replace an older version of software with a newer version immediately, especially if the newer version is significantly different from the older one. Instead, we evaluate the benefits of the newer software and the time cost of retraining. Also, we usually wait to read the reviews of new software and for the "bugs" in the software to be discovered by others and fixed by the manufacturer. After making a decision to change software, we provide our users with both versions during a transition period.
Server Databases. We have entered on the network the title, instructors, and lecture outlines for each course taught in the CVM. Users access the KSU libraries' online catalog (LYNX) and the MEDLINEÆ, CAB, GenBank, and CD Strains databases of the biomedical journal literature. We have an online telephone directory. We have even posted the daily cafeteria lunch menu on the network.
Print Servers. We wanted our users to be able to use any printer on the network from any computer on the network easily. We purchased 7 Hewlett-Packard (HP) LaserJet III printers equipped with 4 MB RAM and the "text equation" and Postscript font cartridges. We used the "HP JetDirect Card for Novell Netware" to attach HP LaserJet printers anywhere on our network. The JetDirect card is placed in the expansion slot of a HP LaserJet printer. The card can be equipped with either a BNC port for connecting to thin coaxial cable or a RJ-45 port for connecting to twisted-pair cable. Connecting printers directly to the network means there are no speed or cable length limitations of parallel-connected printers. This enables files to be sent to the printer at nearly the speed of the network (10Mb/s) instead of the much slower rate of 10 to 30 kilobits (Kb/s) when files have to pass through the printers' parallel port (34). In our laboratories, where we have several printers near one another, we used the Castelle (35) LANpress print server which connects directly to the network cable and supports up to 4 printers operating at network speed. Both of these methods of attaching printers to the network eliminated the need for a computer to be used as a dedicated or semidedicated print server.
Remote Access. Up to four users can simultaneously access all authorized network services including files, application software, network printers, and electronic mail via their off-campus computer and modem. To give faculty, students, staff, and clients access to the network from home, from an office (on- or off-campus) not directly connected to the network, or from a portable computer while traveling, we purchased Microdyne's (36) Netware Access ServerTN and Novell's access server software (for a total cost of $5327). The server provides our users at home with access to the same applications and data that they have at the office at nearly the same level of performance (transmission speed is limited by the speed of the user's modem). We installed four telephone lines equipped with a feature that detects when a line is busy and automatically rotates through the remaining lines until an available one is found ($150 installation, $100/month lease). More remote access ports can be added if necessary. The computer services staff has been the largest user of the remote access server so far. After establishing databases and assigning people to update them, we will encourage clients of the hospital who own a computer and modem to use the remote access server to view their animal's hospital records, send mail messages to faculty, receive information about animal care and other educational material, and receive information about CVM activities. The remote access server enables authorized individuals to reverse the phone charges when calling long distance. It comes with four high speed (up to 38,400 bps) internal modems, which support the latest industry communications standards, including error correction (CCITT V.32 bis and V.42 bis). The remote access server eliminates the need for a dedicated computer for each dial-in user. The server also allows Apple MacIntosh users to run DOS applications across the network remotely or locally. The performance of the remote access server can be monitored locally or remotely by the network manager who can dynamically configure individual ports to improve performance or resolve problems. The software (OnLAN) required by the dial-in user to connect with the remote access server is licensed to the CVM to be copied freely to any number of computers. Password protection, a computer database of authorized users, and the dial-back feature inhibit uncontrolled remote access to the network.
Network Monitoring and Control
To be used, the network must run continuously and smoothly. Therefore, we had to provide ongoing, efficient, and reliable network services. We did not want to rely on users to report poor network performance if it should occur, frustrate users by allowing quality of service to deteriorate, or be driven by user complaints to improve the network. We wanted to be able to anticipate network problems and user needs for network services. We adopted a philosophy of proactive network management and quality control.
The kinds of network problems that might occur are 1) traffic, e.g., when users experience slow response time to their requests for service; 2) access, e.g., when users are denied access to a network device or software application; and 3) erroneous information, e.g., incomplete or garbled information. If traffic is a recurring problem, then 1) the power of the file servers may need to be increased, e.g., by increasing the speed and capacity of the file server's disk drives, RAM, and/or central processor; 2) faster cables may need to used; or 3) the network topology may need to be changed or modified. The effects of short-term traffic can be minimized by the concentrator, which is capable of detecting collisions (when network transmissions "run into" each other and must be transmitted again) and automatically diverting transmissions to another port until the collisions are reduced.
Access to a network device may be denied because of a breakdown of the device or a break in the connection to the device. Network management software alerts the network manager of breakdowns and helps the manager identify the location of broken or bad connections (connectors are more likely to be the source of a problem than the cable).
Access to software may be denied because the software is incompatible with or improperly installed on the network; because the index file used by the network operating system to find the program files has become damaged; or because one or more of the program files or a device driver required by the program has been corrupted. A file is corrupt if information is missing from it, has been improperly added it, or has been garbled. A file can become corrupt from extraneous signals (noise or static), from a power outage or waning, from a "bug" in the software, or from a system crash. The problem is usually corrected by re-installing the network-compatible software.
Network transmission can be corrupt when it loses information (e.g., when a signal has lost strength) or gains information (e.g., from extraneous signals-noise- or from faulty hardware). To assure that signals on the network are transmitted correctly, the hardware has been equipped with sensors to detect improperly sent or received information. Network management software uses information from the sensors to correct many of these problems automatically and to alert the network manager. In the language of computer networks, these sensors detect misalignments, giants, runts, bad CRC's, late collisions, excess collisions, carrier-sense, single collisions, multiple collisions, and other improperly sent or received information. A discussion of each of these is beyond the scope of this paper, but the network manager must use this information about the quality of transmissions to provide continuous quality of service to network users. This does not require much investment in time by the network manager, because so many problems are self-correcting, the network manager is alerted when thresholds are exceeded, and the statistics can be displayed graphically over time. Approximately 1 hour per week is spent on network maintenance.
Network management software alerts the network manager to breakdowns and slowdowns. We purchased SynOptics' LattisNet network management software to monitor day-to-day network performance. The software aids in troubleshooting and problem resolution. If problems occur, the network manager can use the software to restore service quickly, because the manager is alerted quickly that a problem has occurred and can identify the location of the problem. If a file server has failed, we can replace it with another file server loaded with the same software within minutes. If a concentrator port has failed, we can switch the line to another unused port within minutes (some concentrators will do this automatically). If a network card in a workstation has failed, we can replace it with a spare within 30 minutes. If a concentrator module or a print server has failed, we would have to order another one which might take several days to replace, but these network components are the least likely to fail. Even if the entire network should fail, the user's workstation can still be used as a stand-alone computer and there is enough redundancy in computer workstations and printers in the College that the work of our users would not halt.
The network management software can be intuitively used because it graphically displays the network configuration. It enables our network administrator to view a graphic color image of the concentrators and the linkages among them. The concentrators are critical because they are the devices through which the transmissions of most users must pass. The software provides concentrator and port-level performance monitoring, diagnostics, and partitioning (assigning specific paths of transmission within a concentrator instead of allowing the paths to be randomly selected). From a management station running LattisNet Basic Network Management software, the network administrator can monitor and control the concentrators (e.g., transmission paths) and their individual parts (e.g., port assignments and thresholds) in real time and make configuration changes while the network remains up and running. It lets the network administrator quickly locate, isolate, and troubleshoot problems discussed above. The network administrator can selectively observe several concentrators simultaneously to compare performance (e.g., error levels and traffic) at various locations throughout the network, enabling the network manager to locate problems quickly and respond immediately to maintain network performance. The network administrator sets his own fault and activity thresholds and if a threshold is exceeded, an alarm message will alert the network manager. The software reports on network performance and diagnostic data. Graphs of diagnostic data are displayed so the network manager can analyze historical network performance.
Network Reliability. We have minimized the risk of downtime by attaching an uninterruptable power supply (UPS) to each file server; backing up all data files daily onto magnetic tape space (37) and storing the tapes in another building; using the disk mirror option of Novell that automatically copies data files onto another hard disk so that, if failure occurs on one disk, the data are automatically retrieved from the other; and keeping another copy of program and data files on another file server. Isolating user data files from applications on the network and screening software reduces the probability of transmission of computer viruses. Audit trails are kept on-line for detection of unauthorized user access.
Network Antivirus Software. A virus is a program copied to the network by mischievous people that disrupts the network or is destructive to program or data files. A virus is contagious in that it is designed to be transmitted from one computer to another. Currently, we rely on Novell's security to prevent virus infections. We also have experimented with a network virus protection program obtained through shareware, but have found that it interferes with the software we use to back up the file servers that store Apple MacIntosh files. Commercial network antivirus software is expensive ($2000-$3000).
Training. We have found that people learn how to use a computer when they are asked to do so or when they want to do what they have seen others do with a computer. They learn best by doing and asking their peers questions. Therefore, we depend on our users to share what they have learned with others. If a user requests to be trained on the use of a particular software or hardware product, we require them to train others as an exchange of services.
To promote user self-instruction, we have created on-line help and publish a computing news periodical. Users can post questions on the electronic bulletin board or directly to a consultant. We have hired students to be computer consultants. They work in shifts so that one is on duty in our largest computer laboratory at all times between noon and midnight daily. In addition to answering user questions in the laboratory, they are responsible for answering questions over the computer hotline and over electronic mail. They also are responsible for preparing on-line help on topics that our users need (e.g., wordprocessing, spreadsheets, databases). We used a software program called Toolbook to create on-line help.
We conduct computer classes periodically and encourage our users to enroll in computer classes taught on campus. We also require students to complete assignments using a computer for many of the core curriculum courses.
Ease of Use. We have tried to make access to the network and the services it provides as simple as possible for the user. When a user requests a network login account, a set of DOS batch files is created in the login directory of the user's computer. This allows the computer to be used as a stand-alone device or as a device that can be attached to the network; the user decides. Should the network or a user's segment of the network fail, the user still can use the computer stand-alone. User files are automatically stored on the user's computer unless the user elects to store them on a file server. The user has the option of using MicroSoft's Windows, which displays on the user's computer screen the software available to the user on the network. The user makes a selection by moving the cursor to the icon of the application and pressing (clicking) the mouse button.
Network Etiquette. To conserve disk space on the file servers, we ask users to store all files that are not shared with others on their own hard disk, diskettes, or tape. Users that consume large quantities of disk space on the file servers are contacted to see if they have a legitimate need for the disk space or are just not copying their files to their own storage media. The network software allows the network manager to restrict the amount of disk space for repeat abusers. We warn users that the hard drives in the computer laboratories are periodically erased and so should not be used for long-term storage of user files. We also ask that users not overwhelm other users with unnecessary electronic mail.
Network Maintenance. The responsibility for installing and monitoring software and hardware on the network has been distributed among the three computer services personnel. If the network maintenance responsibility were to be assigned to a single person, it would be a full-time job. The responsibilities include installation of new software and hardware, file backup, user instruction, troubleshooting, and performance monitoring and enhancements.
Administration. Our computer staff consists of 4 people: 3 programmers (only 1 of whom received formal training in computer science) and a computing assistant. These people form a committee that meets once a week to coordinate efforts, but each person is housed in a separate area and has a different supervisor. Selection of the 4 areas was based on their computing needs: they are 1) the business office, 2) the hospital, 3) the Practice Management Center, and 4) the Group in Epidemiology, Biostatistics, and Informatics. This distribution of support personnel prevents "empire building" and assures that when users in an area need computing help, they have immediate access to someone. An executive committee composed of the area supervisors and the Dean meets to plan and establish priorities.
Cost. Corporate and alumni funds were used to cover the entire cost of the network infrastructure. The total initial startup cost of building the network infrastructures was $56,301 (Table 1). This provided 216 UTP ports, 6 fiber optic ports, and 8 thin-net ports. When we have had to expand the network, we have done it incrementally 36 ports at a time at a cost of $5353 (Table 2). In addition to the infrastructure costs, each computer to be attached to the network needed a network card costing about $120 . The cost of installing a network card, for copying the software on the user's computer that gives the user access to the network, and attaching an "A-MOD-TAP-B" splitter to the user's telephone cable was about $20. Thus, the total cost (infrastructure and individual station hook-ups) for attaching the initial 200 offices was approximately $450 each. Subsequent connections were approximately $300 each. The network operating system and applications software for the file servers cost approximately $10,000.
The Computer Services Group has an annual budget of $10,000 for supplies, replacement parts, and new purchases. We expect the concentrators to have at least a 5-year lifespan. However, SynOptices reports the measured mean-time between failures for their concentrators to be more than 300,000 hours (>34 years of continuous operation). A port, module, or power supply might fail on a concentrator during this time. For budget purposes, we project the lifespan of the servers and computer workstations to be 2 years, but they will probably be used for 5 years. We try to match the power of the computer to the computing needs of the user. Therefore, we may purchase a more powerful computer than is needed for a new user. The new computer is given to a power-user and the new user receives the power-user's computer. By swapping machines, we can do a better job of keeping up with advances in technology while meeting all of our users' needs.
Future Objectives. Future objectives include placing the hospital record-keeping system on the network, operating laboratory instruments via the network, direct capturing of data from laboratory instruments into network relational databases, accessing multimedia patient records over the network, and transporting interactive multimedia lessons via the network.
The CVM-KSU network has given everyone access to a rich diversity of computing hardware and software that makes them more productive. The network has integrated a variety of computers, workstations, servers, printers, and other devices. Making use of existing, ordinary, unshielded, twisted-pair (UTP) telephone wire in a star topology was cost effective and a flexible method of attaching office computers to the network. Using a thin coaxial cable run to connect computers in laboratories and linking the run to a concentrator increased the number of users without having to add ports to the concentrators and enabled the computer laboratories to be isolated by a bridge if needed to improve network performance. The network supports multiple cable types, network protocols, and operating systems. Its flexibility makes it easily adaptable to future changes. The network has allowed users to share expensive laser printers, plotters, large-capacity disk drives, tape drives, multimedia equipment, and high speed modems. Now software is purchased fetwork use, even though it is more expensive than purchasing software for a single user, because everyone benefits. Multiple purchases of the same software have been eliminated, and all users have the latest version of software, eliminating previous problems of data file incompatibilities. Because software cannot be copied from the network but is accessible to everyone, individual infringement of University software licensing agreements has been reduced, if not eliminated. Powerful management software helps our computer personnel take a proactive approach to network management and growth. The network was designed to accommodate additional users without sacrificing quality of service. The network has been reliable. Since it went on-line, there has been no downtime. The network has been designed so a break in one connection will not bring down the entire system.
Should a file server fail, another file server with identical software can be placed on line within minutes. By purchasing products that conform to industry standards developed by leading standards committees; e.g., Institute of Electrical and Electronics Engineers (IEEE), we have increased our ability to add devices from a variety of manufacturers at competitive prices. The per station cost was nominal. The benefits of the College-wide network have far outweighed those costs. Most importantly, because people in the College have a network to connect them, no one is denied access to information and computing resources.
Table 1. Equipment used in building the computer network for the College of Veterinary Medicine, Kansas State University.
Qty. Equipment Unit Cost Total 1 SynOptics 3000-01 Premises Concentrator $ 2,394 $ 2,394 3 SynOptics 2310-02A Department Concentrator 7,194 21,582 4 Model 504-ST Fiber Optic Concentrator 458 1.832 3 3308 10BASE-T Host Module - (12) port RJ-45 UTP 2,220 6.660 1 3304-ST Ethernet Fiber Optic Host Module - (6) port 2,544 2,544 1 3301 Ethernet Thin-net Host Module - (8) port 1,602 1.602 1 636 Control Console Package for use in an IBM compatible personal
computer with Synoptics- Advanced network management software
4,752 4,752 450 ft. Fiber Optic Cable, Type LGBC-12A 2.34/ft. 1,053 300 ft. Fiber Optic Cable, Type LGBC-6A 1.25/ft. 375 4 AT&T 36 Port Patch Panels 192 768 3 Transceiver Cable 35 105 36 AT&T Fiber ST II Connectors 8 288 73 AT&T 2.5' Patch Cords 4 292 7 AT&T Dual Fiber Patch Cords 91 637 4 AT&T LightGuide Interface Unit w/ Connector Panels 68 272 36 AT&T ST Connector Couplings 8 288 4 Wall Racks (19") 146 584 8 Mounting Brackets for Blocks 37 296 LABOR 4,877 TOTAL $56,301
Table 2. Cost for adding 36 ports to the computer network for the College of Veterinary Medicine, Kansas State University.
Qty. Equipment Unit Cost Total 1 Synoptic 2310-02A Department Concentrator $5,124 $5,124 1 36 port Ortronics OR808044324 Telco Patch Panel for use with 2 pair
punch-down wires and RF-45 phone connectors
107.00 107 36 Ortronics 8010 DTP8003DE 3 ft. Patch Cables, Level 3 stranded
3.39 122 TOTAL $5,353
Our telecommunications department charged $500 to wire the concentrator to the telephone wire (i.e., wire the punchdown block to the patch panel of the concentrator) and $10 per line for a "A-MOD-TAP-B" Y splitter in each office that was connected adding $860 to the total cost.
Figure 1. (provided by author) An unshielded twisted-pair patch cable with RJ-45 connectors on each end extends from the concentrator to a patch panel. A patch panel tie cable connects the patch panel to a punch-down block, where unshielded twisted-pair wire is distributed to individual phone jacks in offices and laboratories. Thin coaxial cable is used to attach workstations in computer laboratories and file servers to the network. Concentrators are connected by a fiber optic cable (backbone). A bridge connects the College network to the rest of the campus and confines local network traffic from being broadcast throughout the campus.
References and Endnotes
1. The model number and cost of the products are listed only as a reference guide to the specifications of the network components and do not constitute an endorsement by Kansas State University. The products were purchased by competitive bid. Other manufacturers may produce equivalent or better equipment, which should be evaluated by potential purchases.
2. IBM Corportation, Bldg. 983, 11400 Burnett Road, Austin, TX 79758.
3. Apple Computer, Inc., 20525 Mariani Ave., Cupertino, CA 95014.
4. Digital Equipment Corporation, 146 Main St., Maynard, MA 01754.
5. Sun Microsystems, 255 Garcia Ave., Mountain View, CA 94043.
6. Southwestern Bell Telephone Co., P.O. Box 940012, Dallas, TX 75394-0012.
7. SynOptics Communications, Inc., 501 East Middlefield Road, Mountainview, CA 94043-4015.
8. Worry-Free 10BASE-T, 3Com Corporation, 1991, TECHDOC, 3676.
9. Novell NetWare, Novell, Inc., 122 East, 1700 South, Provo, UT 84606.
10. Banyan's VINES, 120 Flanders Road, West Borough, MA 01581.
11. IBM Token Ring, IBM, Old Orchard Road, Armonk, NY 10504.
12. 3Com Corporation, 5400 Bayfront Plaza, Santa Clara, CA 95052.
13. LAN Manager, Microsoft Corporation, Microsoft Way, Redmond, WA 98052.
14. Artisoft, Inc., Artisoft Plaza, 575 East River Road, Tucson, AZ 85704.
15. Appletalk, Apple Computer, Inc., 20525 Mariani Ave., Cupertino, CA 95014.
16. Cisco, AGS+; Cisco Systems, 1525 O'Brien Drive, Menlo Park, CA 94025.
17. Kansas Communications, 660 North Lindenwood, Olathe, KS 66062.
18. Cabletron, 35 Industrial Way, Rochester, NH 03878.
19. 3Com Corporation, 5400 Bayfront Plaza, Santa Clara, CA 95052.
20. Fibermux Corporation, 9310 Topanga Canyon Blvd., Chatsworth, CA 91311.
21. Clarkson University, Potsdam, NY 13699.
22. Microtech Computers, 2329 Iowa Street, Suite M, Lawrence, KS 66046.
24. RBBS. Shareware, Splicer.cba.hawaii.edu.
25. Lotus 123, Lotus, 55 Cambridge Parkway, Cambridge, MA 02142.
26. DBASE IV, LAN Version 1.1., Ashton Tate, 20101 Hamilton Ave., P.O. Box 2833, Torrance, CA 90509-2833.
27. SAS, SAS Institute, Inc., SAS Circle, Box 8000, Cary, NC 27512-8000.
28. BMDP Statistical Software, Inc., 1440 Sepuveda Blvd., Suite 316, Los Angeles, CA 90025.
29. Systat, Version 4.1, Systat, Inc., 1800 Sherman Ave., Evanston, IL 60201.
30. EGRET, Version 0.26.6, Statistics and Epidemiology Research Corp., 1107 NE 45th, Suite 520, Seattle, WA 98105.
31. EpiInfo, USD, Inc., 2075A West Park Place, Stone Mountain, GA 30087.
32. SigmaPlot, Jandel Scientific, 65 Koch Road, Corte Madera, CA 94925.
33. Reference Manager, Research Information Systems, Inc., Camino Corporate Center, 2355 Camino Via Roble, Carlsbad, CA 92009.
34. Connecting with the future of network printing. Hewlett-Packard Co., 1992, 5091-3527EUS.
35. Castelle, Inc., 3255-3 Scott Blvd., Santa Clara, CA 950534
36. Microdyne, 207 South Peyton Street, Alexandria, VA 22314.
37. Mountain Network Solutions, 240 East Hacienda Ave., Campbell, CA 95008-6623.
The author wishes to acknowledge Michael Lorenz who acquired the funding for networking the College: Hill's Pet Products and KSU-CVM alumni for providing those funds; Robert Hetrick, Pat Oblander, Gregg Knittel, Terry Teske, and Sandi Allard, our computer services staff who were instrumental in implementing the network; Jerry R. Gillespie, Collis P. Bosworth III, Jack Mara, David Geier, Rick Summerhill, Doug Elcock, and David Leith who provided advice on the network.
Address-A name, set of numbers, or sequence of bits used to identify devices (computer, printer, or server) on a network.
Backbone-A solid length of cable extending across the entire network to which network nodes are attached along its length.
Bandwidth-The capacity of a network to carry information using a particular type of cable (or medium), as measured by the maximum number of bits per second (bps) the network can transmit, typically 10 MegaBits per second.
Bridge-An electronic device that connects two networks so that devices on one network can communicate with devices on the other network.
Client-A device on the network such as a user's computer, a printer, a modem, or another server.
Coaxial Cable-An electrical cable with an insulated solid wire inside an insulated braided-mesh wire tube. Both wires have the same center point, or axis, which is why the cable is named coaxial. The solid wire carries data, whereas the tube acts as a shield for the solid wire. Coaxial cables used for networks usually are of two diameters, thin (called thin-net) and thick (called thick-net).
Collision-When network transmissions "run into" each other and must be transmitted again.
Concentrator-An "intelligent" wiring hub with information-processing capability.
Driver-A disk file that contains the information needed by a program to operate a device.
Ethernet-A local area network standard originally developed by Xerox. Ethernet networks normally use twisted-pair cabling and coaxial cables of two types, thick or thin. The transmission scheme is baseband at a data rate of 10 Mbits-per-second.
Fiber Optic Interface-A box usually made of metal and wall-mounted that provides a place to secure the ends of fiber optic cable and any slack in the cable.
Gateway-An electronic device that connects two networks, each of which operates with a different set of protocols. The gateway translates all the protocols of one network into those of the other network, so workstation and other devices on the two networks can communicate with one another.
LAN (Local Area Network)-A network in one location. The size of the location can vary; for example, a network on one floor of an office building is an LAN. So is a network connecting all the buildings of a college campus. Networks connected by modems and telephone lines are not LAN's.
Mail Server-A computer on a network that acts as an electronic post office. People using the network transmit a message to the mail server, where the message is stored until the addressee checks the server and reads the message.
Modules (Host Modules)-Electronic boards (cards) that fit into a concentrator and serve a particular function such as network management of connecting wire of a specific type to the concentrator.
Network-A collection of individually controlled computers, printers, modems, and other electronic devices interconnected. Networks also include all the software to run the network. Using a network, people can share date, programs, and send messages to each other. Networks also let people share resources, such as printers and storage units.
Network Card (Network Interface Controller)-A card (or set of chips) that fits inside a computer so that the computer can connect to a network. The card has a connector for attaching the network cable.
Node-Any device that has an address o a network such as a workstation, printer, or file server (modems and star hubs usually are not nodes).
Patch Panel-A board to which connectors of one type have been attached on one side and connectors of another type have been attached to the other side.
Patch Cable-A short length of cable used to connect wiring devices such as a concentrator to a patch panel.
Punch-down Block-A block, usually metal, that has connectors for telephone wire. To make a connection, the wire is pressed down (punched) between two metal posts by use of a hand tool.
Print Server-A device (usually an additional workstation) that acts as a large buffer for files being sent to the printer. Also called a spooler. Allows users to print at the same time they are working on some other task. When users send data to the printer, the data are sent to the server and stored there until the printer is available. Then, the print server automatically sends the data to the printer to be printed.
Protocol-A procedural rule or convention for sending information over a network. Each network has its own way of transmitting data and divides the entire process into a series of specific functions. Accomplishing each function requires a complete set of operating rules or protocols. The rules describe the content, format, timing, and error control of messages.
Repeater-In local area networks, a hardware device used to extend the length of network cabling by amplifying and passing along messages traveling through the network. Repeaters are necessary because, as signals pass over a line, they lose some of their power and pick up static.
Ring Topology-A network arrangement in which all the devices are connected in a circle or ring. Data passes around the ring from node to node always in the same direction. Each node acts essentially as a repeater and retransmits the messages to the next node.
Router-A device that connects two networks together by maintaining addressing information for each network and prevents the local communications of one network from traveling throughout the other network.
Server-A computer that serves files to other computers and devices on the network. The three most common types of servers on local area networks are the print server, file server, and mail server.
Shielded Cable-A wire or circuit enclosed by a grounded metallic material. Shielding serves two purposes. First, it keeps outside electrical disturbances from reaching the wire and disrupting the signals passing over it. Second, shielding keeps the cable from emitting radiation.
Star Topology-A centralized network with all the nodes connected to the main computer via a phone line at a punch-down block in the closet.
10BASE-T-The Institute of Electrical and Electronics Engineers (IEEE) specification that defines the type of cards, connectors, and hubs required to put ethernet onto twisted-pair wire.
TCP/IP (Transmission Control Protocol/Internet Protocol)-A set of protocols designed to let many networks interconnect. TCP/IP is the standard for internetwork communication established on the U.S. Department of Defense network known as ARPAnet. ARPA stands for the Advanced Research Projects Agency. It is the research and development arm of the Department of Defense (and is also called DRPA). TCP/IP is rapidly becoming a de facto standard for network interconnnections for universities and research organizations. It has been associated with UNIX networks because various companies selling UNIX devices have built the TCP/IP protocols in the devices.
Topology-The physical layout of a network, including the cables and devices. The topology of a network is its "roadmap" and can show which devices can communicate directly and which cannot because they do not share a connection path.
Transceiver-Hardware that sends and receives network signals.
Twisted-pair Cable-Ordinary telephone wire consisting of two insulated copper strands twisted about each other to reduce outside interference of their signals. Sometime referred to as telephone twisted-pair or TTP. Twisted-pair wiring is relatively inexpensive, easy to install, easy to modify, and may already exist in many installations as part of the telephone network.