As mentioned in Chapter 2, "Advantages of Client/Server Computing," the cost of powerful hardware for client/server computing has declined dramatically in the last few years. Nevertheless, this
power must be packaged properly, and cost still must be considered in the design and purchasing decision. Hardware that provides the client/server, LAN-to-LAN, and LAN-to-WAN connectivity must be acquired for clients, servers, data storage, and the
Entry-level client workstations can range from a basic Intel-based PC to an entry-level Apple Macintosh or an X-Terminal. These entry-level clients start at about $1,000 and use LAN servers for printing, backup, software storage, application execution,
and WAN connectivity. High-end client workstations can cost more than $50,000 for engineering stations that provide advanced capabilities such as a gigabyte or more of local storage, high-resolution graphics monitors, 100-MIPS processing, direct WAN
connectivity, 1000-dpi color printing, or professional multimedia development tools. The average client workstation has dropped from $5000 to $2000 in the last two years. This buys a configuration with the processing power equivalent to an 8Mbyte Intel
33-MHz 486DX PC with local storage of 250Mbytes, LAN connectivity, and a VGA-equivalent monitor. This cost level is not expected to decline much further, because GUI software and reengineered application requirements will steadily increase the processing
power requirements for entry-level machines.
Server hardware offers the largest and most complex set of choices. Servers run the gamut from a $30M+ traditional IBM mainframe, to a 4- to 16-way symmetric segment multiprocessor machine, to a 32- to 32767-processor massively parallel cluster
supporting hundreds of users, to a $5,000 PC used to provide file and connectivity services for a small LAN workgroup. Many organizations also have client/server applications that use the services of existing IBM 370 mainframes running VM, MVS, or VSE, DEC
VAX minicomputers running VMS or Ultrix, and large RISC-based systems running UNIXall as high-end servers.
Other mainframe and minicomputer hardware platforms, running proprietary operating systems, are frequently used in terminal emulation mode from the client workstation. The non-IBM and DEC proprietary operating system platforms rarely are used to
provide other services, such as database and RPC-invoked application services. There is a lack of tools available in these environments to build or buy client/server applications. Servers based on the IBM, DEC, and UNIX operating systems will provide
application services using existing applications through terminal emulation or RPC-invoked application services. These same servers will provide connectivity and database services to the first client/server applications in an organization.
Connectivity requires every client workstation to be connected to a LAN or through a WAN to a remote server. In the usual situation, the workstation is connected through an Ethernet, Token Ring, FDDI, CDDI, or occasionally a parallel or serial
interface to the LAN. The primary connection types require a network interface card (NIC) to be inserted in the workstation to provide the protocol processing necessary to establish and maintain the connection. The cost of LAN connectivity has declined
rapidly in parallel with the industry reduction in workstation costs.
Cabling costs vary widely, depending on the physical difficulty of installation and whether the network planners choose unshielded twisted-pair (UTP), shielded twisted-pair (STP), or glass-fiber cables. Cable costs without installation run from $1 per
foot for UTP, $1.50 per foot for STP, to $3 per foot for glass fiber. Installation costs vary from $1 per foot to $15 per foot, depending on the physical environment and connection requirements. Glass-fiber termination equipment is more costly than
twisted-pair, although the costs are declining. Current costs are between $100-200 for Ethernet, $300-500 for Token Ring, $300-700 for CDDI, and $750-1250 for FDDI.
Today, many vendors provide the hardware for these connections. Each vendor offers some advantages in terms of cost, performance, and reliability. Motorola provides wireless Ethernet connectivity at lower speeds and higher costs than wired connections.
Wireless connections are an advantage in existing buildings with no cable installed and with relatively low-speed communications requirements.
WAN connectivity requires each workstation to be directly connected to the WAN or to a communications server connected to the WAN. Most new LANs are installed using communications servers. There are cost, performance, and especially network management
reasons for using a LAN communications server. A substantial advantage accrues because there is no need to cable each workstation to the WAN. Workstations that are individually connected to the WAN require an embedded controller card for synchronous
communications and either a modem or serial connection for asynchronous communications. These typically operate at speeds of 2400-64000 bits per second (bps) through analog or digital modems. Each workstation must have its own cable connecting it to the
WAN controller. Workstations connected to the WAN through a communications server share a higher-speed connection, typically 14400 bps, 56000 bps, or 1.54 Mbps.
A major advantage of the communications server is its ability to multiplex a high-speed communications line and provide bandwidth on demand to each client workstation. Only the single LAN cable and LAN controller are needed to provide workstation
connectivity in the server implementation.
Data storage can be provided to a client from a local disk or through the file services of the NOS. Local disk storage requires the workstation to have its own disk devices. Server storage involves large shared server disks. In either case, a backup
capability must be provided. This can be done through local diskette or tape devices or though a server tape, disk, or optical device.
Before selecting client hardware for end users, most organizations should define standards for classes of users. This set of standards simplifies the selection of the appropriate client hardware for a user and allows buyers to arrange purchasing
agreements to gain volume pricing discounts.
There are a number of issues to consider when selecting the client workstation, including processor type, coprocessor capability, internal bus structure, size of the base unit, and so on. However, of these issues, one of the most overlooked regarding
client/server applications is the use of a GUI. GUI applications require VGA or better screen drivers. Screens, larger than the 15-inch PC standard, are required for users who normally display several active windows at one time; the more windows active
on-screen, the larger the monitor viewing area requirements. The use of image, graphics, or full-motion video requires a large screen with very high resolution for regular usage. It is important to remember that productivity is dramatically affected by
inability to easily read the screen all day. Inappropriate resolution will lead to fatigue and inefficiency.
The enterprise on the desk requires that adequate bandwidth be available to provide responsiveness to the desktop user. If regular access to off LAN data is required, a router based internetworking implementation will be required. If only occasional
off LAN access is required, bridges can be used. Routers provide the additional advantage of support for multiprotocol internetworking. This is frequently necessary as organizations install 10BaseT Ethernet into an existing Token Ring environment. Fast
Ethernet and FDDI are becoming more prevalent as multimedia applications are delivered.
Client/server applications vary considerably in their client processing requirements and their I/O demands on the client processor and server. In general, clients that support protected-mode addressing should be purchased. This implies the use of
32-bit processorsperhaps with a 16-bit I/O bus if the I/O requirement is low. Low means the client isn't required to send and receive large amounts of data, such as images, which could be 100K bytes or larger, on a constant basis.
As multiwindowed and multimedia applications become prevalent during 1994, many applications will require the bandwidth only provided by a 32-bit I/O bus using VESA VL-bus or Intel PCI technology. Windowed applications require considerable processing
power to provide adequate response levels. The introduction of application integration via DCE, OLE, and DOE significantly increases the process ing requirements at the desktop. The recommended minimum configuration for desktop processors has the
processing capacity of a 33Mhz Intel 486SX. By early 1995, the minimum requirement will be the processing capacity of a 50Mhz Intel 486DX or a 33Mhz Intel Pentium.
The Mac System 7 operating system is visually intuitive and provides the best productivity when response time to GUI operations is secondary. The Motorola 68040, 8Mbytes RAM, 120Mbyte disk is recommended. By early 1995, the availability of PowerPC
technology and the integration of System 7 with AIX and Windows means that users will need considerably more processor capacity. Fortunately, the PowerPC will provide this for the same or lower cost than the existing Motorola technology.
Users working remotely on a regular basis may find that a notebook computer best satisfies their requirements. The notebook computer is the fastest growing market today. The current technology in this area is available for Intel PC, Apple Macintosh,
and SPARC UNIX processors. Because notebooks are "miniaturized," their disk drives are often not comparable to full-size desktop units. Thus, the relatively slower speed of disk I/O on notebooks makes it preferable to install extra RAM, creating
"virtual" disk drives.
A minimal configuration is a processor with the equivalent processing power of a 33Mhz Intel 486SX, 8mbytes of RAM and 140Mbytes of disk. In addition, the notebook with battery should weigh less than seven pounds and have a battery life of three hours.
Color support is an option during 1994 but will be mandatory for all by 1995. In addition, if the application will run a remote GUI, it is desirable to install software to compress the GUI and V.32 modem communications at 9600 bps or V.32bis at 14400 bps,
employing V.42 and V.42bis compression, respectively. The effective throughput is two to three times the baud rate because of compression. The use of MNP4 and V.42 or MNP5 and V.42bis error correction enables these speeds to work effectively even during
noisy line conditions. The introduction of PCMCIA technology, credit card size modems, and flash memory are available to upgrade the notebook.
Pen-based clients provide the capability to operate applications using a pen to point and select or write without the need for a mouse or keyboard. Frequently, they are used for verification, selection, and inquiry applications where selection lists
are available. Developers using this technology use object-oriented software techniques that are RAM-intensive.
The introduction of personal digital assistant (PDA) technology in 1993 has opened the market to pocket size computing. During 1994, this technology will mature with increased storage capacity through cheaper, denser RAM and flash memory technology.
The screen resolution will improve, and applications will be developed that are not dependent upon cursive writing recognition.
The PDA market is price-sensitive to a $500-$1000 device with the capability to run a Windows-like operating environment in 4MB of RAM, a 20Mhz Intel 486SX processor, and 8MB of flash memory. Devices with this capability will appear in 1994, and
significant applications beyond personal diaries will be in use. During 1995, 16MB of RAM and 32MB of flash memory will begin to appear, enabling these devices to reach a mass market beyond 1996. In combination with wireless technology advances, this will
become the personal information source for electronic news, magazines, books, and so on. Your electronic Personal Wall Street Journal will follow you for access on your PDA.
UNIX client workstations are used when the client processing needs are intensive. In many applications requiring UNIX, X-terminals connected to a UNIX presentation server will be the clients of choice. A UNIX client workstation will then have more
processing power than a PC client.
The introduction of software from SunSoft, Insignia Solutions, and Locus Computing that supports the execution of DOS and Windows 3.x applications in a UNIX window makes the UNIX desktop available to users requiring software from both environments. The
PowerPC and Sparc technologies will battle for this marketplace. Both are expected to gain market share from Intel during and after 1994.
X-terminals provide the capability to perform only presentation services at the workstation. Processing services are provided by another UNIX, Windows 3.x, NT, OS/2 2.x, or VMS server. Database, communications, and applications services are provided by
the same or other servers in the network. The minimum memory configuration requirement for an X-terminal used in a client/server application is 4-8 Mbytes RAM, depending on the number of open windows.
Server requirements vary according to the complexity of the application and the distribution of work. Because servers are multiuser devices, the number of active users is also a major sizing factor. Servers that provide for 32-bit preemptive
multitasking operating systems with storage protection are preferred in the multiuser environment.
Intel-based tower PCs and Symmetric Multi-Processors (SMPs) are commonly used for workgroup LANs with file and application service requirements. Most PC vendors provide a 66Mhz Intel 486DX or Intel Pentium for this market in 1994. SMP products are
provided by vendors such as IBM, Compaq, and NetFrame. Traditional UNIX vendors, such as Sun, HP, IBM, and Pyramid provide server hardware for applications requiring UNIX stability and capacity for database and application servers and large workgroup file
The SMP products, in conjunction with RAID disk technology, can be configured to provide mainframe level reliability for client/server applications. It is critical that the server be architected as part of the systems management support strategy to
achieve this reliability.
Permanent storage requirements are very application-specific. In addition to quantity of disk storage, the issues of performance and reliability must be considered.
Disk storage devices should use the SCSI-2 standard controller interface. This provides the best performance in a standards-based environment. Many vendors provide high-capacity, high-performance, and highly reliable disk devices for this controller.
The use of high-capacity cache storage dramatically improves performance. Most current SCSI-2 controllers are configurable with 256K or more cache. This is an important, yet frequently overlooked component of the architecture. New drives are available
in the traditional 3.5 size with 1.0-1.6Gbyte capacity. The use of compression software can easily double this capacity. With the increasing size of GUI software and the introduction of multimedia applications, the demand for disk capacity will increase
rapidly during 1994 and beyond.
When applications require high reliability, it may be appropriate to use a configuration that supports mirrored disks. With this configuration, data is automatically written to two disks. This enables the application to continue if a failure occur on
one disk. System files and data files should be considered for mirroring. Even though system files are usually read-only, the number of users affected by unavailability of the files may justify this redundancy. In addition, performance can improve since
dual reads can be handled in parallel.
Traditional magnetic disk technology is often referred to as single large expensive disk (SLED). Very high performance and high availability can be achieved through a redundant array of inexpensive drives (RAID). These enable data files to be spread
across several physical drives. Data also can be mirrored as part of this configuration. RAID technology provides a considerable performance advantage because many parallel I/O operations can be processed at the same time. High capacity caches must be used
in conjunction with RAID to achieve optimum performance. The size will be identified as part of the architecture definition.
Although most permanently stored data uses disk, tape is a very popular form of low-cost magnetic storage and is used primarily for backup purposes. The standard backup tape device today is digital audio tape (DAT). These tapes provide approximately
1.2 Gbytes of storage on a standard cartridge-size cassette tape. Tape is a sequential medium and does not adequately support direct (random) access to information. If an organization standardizes on a single tape format and technology, distribution of
information by mailing tapes can be a cost-effective communications mechanism for large quantities of information that do not require real-time transmission or accessibility.
Optical disk storage technology provides the advantage of high-volume, economical storage with somewhat slower access times than traditional magnetic disk storage.
Compact disk-read only memory (CD-ROM) optical drives are used for storage of information that is distributed for read-only use. A single CD-ROM can hold up to 800MB of information. Software and large reports distributed to a large number of users are
good candidates for this media. CD-ROM also is more reliable for shipping and distribution than magnetic disks or tapes.
By 1995, it is expected that all software and documentation will be distributed only on CD-ROM. The advent of multimedia applications and the resulting storage requirements will further drive the demand for CD-ROM.
In 1993, the speed of CD-ROM technology was doubled through a doubling of the rotation of the drive. Newer drives will triple-spin and quad-spin. The speed of the drive is very critical for applications that use the CD-ROM interactively. The addition
of large cache SCSI-2 controllers can also significantly improve performance. The architecture definition must look at the business requirement in determining the appropriate configuration. Poor selection will result in unacceptable performance, excessive
cost, or both.
Write once, read many (WORM) optical drives are used to store information that is to be written to disk just once but read many times. This type of storage is frequently used to archive data that should not be modified. Traffic tickets issued by police
departments are scanned and stored on WORM drives for reference on payment or nonpayment. The WORM technology guarantees that the image cannot be tampered with. A magnetic drive can be used to store an index into the data on the WORM drive. Data can be
effectively erased from a WORM by removing reference to it from the index. This can be useful when a permanent audit trail of changes is required.
Erasable optical drives are used as an alternative to standard magnetic disk drives when speed of access is not important and the volume of data stored is large. They can be used for image, multimedia, backup, or high-volume, low-activity storage.
Client and server processors are attached to the LAN through NICs. These provide the physical connectivity to the wire and the protocol support to send/receive messages. As discussed in Chapter 5, "Components of Client/Server
ApplicationsConnectivity," the most popular network protocols today are Token Ring, Ethernet, and FDDI. The following paragraphs illustrate key selection issues regarding each technology.
Token Ring NICs were originally IBM-only products but are now provided and supported by many PC and UNIX vendors. The IEEE standard 802.5 defines the standards for the interface. Token Ring NICs are particularly desirable for LANs that are collocated
with an IBM mainframe. They are also useful when interactive use is combined on the same LAN with high-volume file transfer or print image communications. Token Ring LANs operate at 4 or 16 Mbps. Shielded twisted-pair (STP) (Type 1 cabling) is required by
some vendors for 16-Mbps processing, but unshielded twisted-pair (UTP) cable is supported by many at 16 Mbps and all at 4 Mbps.
The rapid decline in price for 10BaseT Ethernet and the increasing availability of Fast Ethernet means that despite some technical advantages the future of Token Ring is limited.
The existing de facto standard for LAN connection defined by the IEEE standard 802.3, Ethernet is supported by almost every vendor. The large number of vendors providing NICs ensures their competitive pricing. Ethernet works well when interactive-only
or file transfer-only communications are present on the LAN. When mixing interactive and file transfer on the same Ethernet system, performance is excellent when LAN loading does not exceed 30 percent of the capacity. Most Ethernet LANs operate at 10 Mbps.
The present standard for Ethernet connectivity 10BaseT operates on STP or UTP. Recent products supporting Fast Ethernet and ATM will provide support for 100Mbit and up to 2.4Gbit on existing Type 5 UTP-cabled network.
Fiber Distributed Data Interchange (FDDI) is a protocol originally defined for high-speed communications over glass fiber. FDDI provides 100-Mbps throughput today. NICs for FDDI are becoming available for more processing environments. This throughput
is necessary when applications deal with large images, large file transfers, or multimedia using full-motion video. The rapid advances in Fast Ethernet and ATM means that FDDI will see limited rollout except for building internetworking and WANs.
Copper Distributed Data Interchange (CDDI) provides support for FDDI communications over copper wire. The same 100-Mbps throughput is supported over Type 1 cabling (STP), and standards are emerging to provide support over Type 3 cabling (UTP) that is
carefully selected and installed. This technology is now called Fast Ethernet. ATM is discussed in Chapter 5 and will increasingly be the protocol of choice for LAN/WAN internetworking.
A lot has been written in books, magazines, and journals about computer hardware and software; and a number of computer specialty businesses are dedicated to helping you work through issues of specific concern to your business objectives. Rather than
cover the minute details here, this chapter has attempted to highlight a number of areas for you.
However, before closing this chapter, one critical area often overlooked (but is the cause of many serious problems when neglected) is power protection.
Prices for UPS have declined to the point where they are widely used for LAN server protection. These units contain battery backups to enable at least a graceful power-down sequence. All buffers can be purged and files closed so that no information is
lost. Other units provide 15-90 minutes of power backup to handle most power failure situations.
The electronics in computer systems are affected by power fluctuations. Protection can be obtained through the use of surge protection equipment. Every client and server processor, and all peripherals should be wired through a surge protector. Most UPS systems include integrated surge protectors.