Infrastructure

Table of Contents

Background

Arizona schools face challenges, similar to inter-state districts and schools, in the knowledge, selection, design, deployment, funding, and maintenance of network infrastructure systems. This section will address these challenges and provide a workable reference for educators who are responsible for making these decisions.

Cabling Systems Evolution

Premises cabling systems for telecommunications services have undergone a constant evolution over the years. Telephone cabling systems were once specified and installed by telephone companies while data cabling was dictated by computer hardware vendors. After the breakup of AT&T, attempts were made to simplify cabling by introducing a more universal approach. While these systems helped define cabling issues, it wasn’t until 1991 when the ANSI/EIA/TIA-568 building wiring standard was published that complete specifications became available to guide cable system selection and installation.

Networking: the Classroom for the Future

Networking is easily the most talked about, yet least understood, technology of the 1990s. A network can be thought of as an information highway, which transports any combination of voice, video or data traffic to its destination. When it is incorporated in the classroom, networking serves as the basis for an innovative, effective and dynamic educational tool. Now, in addition to chalkboard and textbooks, children can be challenged in an interactive computer lab and inspired by remote learning, just two of a myriad of possibilities available with today's technologies.

Data Networking

A data highway is made up of residential streets known as Local Area Networks (LANs) and interstates known as Wide Area Networks (WANs). LANs and WANs can also be interconnected to form a system of roads to travel coast to coast. A data network is composed of five basic components:

• Wiring Media: the wire and cable that transmit data signals

• Networking Products: the electronics for a IAN from workstation equipment (whether it is located in the classroom, computer lab or administrative office) to multi-access method hubs

Each of these components plays a key role in paving a data highway.

The document entitled ‘K-12 Internetworking Guidelines’ (RFC 1709) by Joan Gargano of the University of California, Davis and David Wasley of the University of California, Berkley provides a valuable reference for K-12 information system decision makers. Portions of RFC 1709 are paraphrased in the following divisions of this document.

Existing Standards

1. Physical Infrastructure Development

Premise: Current physical infrastructure development supports the need for systems to be general, flexible and adaptable to future technology advances.

The element that allows for slow or high-speed transmission of data through a district or campus network is the network bandwidth. If the available bandwidth is too small, congestion results. Improvements can be made to increase the bandwidth of network electronics; however, the physical cable has a fixed bandwidth. If it is insufficient, changing the electronics will not help—that is, except for fiber optic cable. Despite its mundaneness, the cable plant should be considered the keystone of the network.

 

Adopt Structured Wiring Guidelines

When planning a structured cabling system, there will be many factors to consider. The bottom line is that the school or district office needs a structured cabling system that's flexible, manageable, cost effective and, at the same time, able to meet education’s performance requirements.⁴ A structured cabling system should be "open," or capable of supporting all desired physical environments, applications and performance requirements. How each district achieves this goal can be the most critical decision.

A structured cabling system consists of a flexible cabling infrastructure that can support multiple computer and telephone systems independent of their manufacturer. In a structured cabling system, each workstation is wired to a central point using a star topology, facilitating system interconnection and administration. This approach allows communication with virtually any device, anywhere at any time. A well-designed cabling plan may include several independent cabling solutions of different media types, installed at each workstation to support multiple system performance requirements.

Structured Cabling Systems - The Basic Building Blocks ⁵

Patch Cable Assemblies

Patch cable assemblies are connectorized cables which interconnect patch panel ports and/or attach workstation equipment to information outlets. Patch cable assemblies make moves, adds and changes quick and easy.

Information Outlets

Information outlets are the point of termination for cable at or near the workstation. They are categorized by their physical environment (flush mount, surface mount, modular furniture, raised floor or poke-through), the number of ports per outlet and the connector types required.

Horizontal Cable

Horizontal cable provides the medium over which communication services are transmitted. Horizontal cable can consist of unshielded twisted pair (UTP), shielded twisted pair (STP) and/or fiber optic cable. Each medium has distinct electrical properties and unique application capabilities (see Performance and Bandwidth Requirements section).

 

Cross-connect Products

Cross-connect products provide a means of termination for cabling while establishing a field for administering moves, adds and changes. There are two types of cross-connect equipment: patch panels and punch-down blocks.

Backbone Cable

A structured cabling system is made up of independent horizontal distribution cables connected via cross-connect products to riser or backbone cabling. The backbone originates at the main distribution point and interconnects with all telecommunications closets. Backbone cables are typically constructed of optical fiber or multi-pair copper.

Product Features

There are a wide variety of product issues to consider when selecting a structured cabling system. These issues involve everything from how the components physically fit into certain environments to the particular features a product offers.

Physical/Environmental Requirements

Products can be selected to accommodate various physical requirements such as rack or cabinet mounting, modular furniture or raised floor environments.

Cross-connect Choices

Cross-connect equipment can vary widely depending upon media type, ease of use and size requirements.

Labeling/Color Coding

Cabling system administration can be greatly aided through the use of colored cables and connecting hardware featuring colored icons or labeling.

Termination Styles

A variety of termination styles are available depending upon media choice and connecting hardware selection.

Media

Plenum or non-plenum cables can be selected featuring one or a combination of media types under one jacket. The choice of a network transmission media for a specific application should depend upon the application that it will support. The factors to be considered include the:

• Flexibility of the medium with respect to supported services.

• Required useful life of backbone wiring.

• School size and student/user population.

Telecommunications service needs of a school building’s occupants may vary as time passes and occupants change. Future uses of the backbone wiring may range from highly predictable to very unpredictable. Whenever possible, determine the different service requirements first. It is often convenient to group similar services together in categories such as voice, display terminal, Local Area Network (LAN), and other digital connections. Then identify the individual media types and projected quantities required within each group.

When requirements are uncertain, use worst-case estimates to evaluate backbone wiring alternatives. The more uncertain the requirements are, the more flexible the backbone wiring system must be.

Each recognized cable has individual characteristics that make it useful in a variety of situations. In some situations, a single cable type may not satisfy all the user requirements. In these cases, use more than one medium in the backbone wiring. The different media should use the same facility architecture with the same locations for cross-connects, mechanical terminations, inter-building entrance facilities, and other facilities.⁶

The cabling system should outlive most data hardware. The new EIA/TIA specifications will likely influence the standards for the next generation of high-speed LANs such as Twisted Pair-Physical Medium Dependent (TP-PMD) and Asynchronous Transfer Mode (ATM). The performance margins offered by the various structured cabling Systems are absolutely critical.

The following graphic diagrams the components of a structured wiring system. (Anixter)

 

 

The shaded area represents all the interconnect products considered part of the "Structured Wiring System." In this example, the workstation side of the overall system uses a patch panel cross-connect. A patch cable is used to patch into any of the system side access methods (e.g., an Ethernet concentrator or a Token Ring MAU). The important concept is that every information outlet has a corresponding cross-connect port in the wiring closet that is used to patch into particular system side equipment for data, voice or video.

In addition to a structured wiring system, a LAN is made up of networking products known as active electronics. These products direct and manage the traffic on the network. Although networking products differ greatly in function - from forwarding and adapting to repeating and reclocking an electronic signal - they all enhance the capabilities of a data network. Some of the products which fall under the networking umbrella include: concentrators, hubs, transceivers, repeaters, network management and other software, and interface cards.

An access method is the engine that drives the information packet along the highway. It allows network devices to access a shared transport media. An access method turns a single lane data highway into a multiple-lane highway by allowing several users to drive or communicate at the same time. There are three primary access methods available today, each providing different features and benefits:

• Token Ring

• Ethernet

• LocalTalk

Token Ring

Most commonly associated with IBM, Token Ring is an access method which communicates with other devices on the network by constantly circulating a "token" on the network's highway. When a user wants to drive on the highway, he must simply acquire the token to communicate. The Token Ring highway can be paved with STP or UTP cable.

Token Ring operates at two speeds, 4 and 16 megabits per second (Mbps) - depending on the level of traffic on your network. As many as 260 drivers or devices can be on a Token Ring network using STP cabling, given certain length restraints.

 

Ethernet

Another access method available is Ethernet. Jointly developed by Xerox and Digital, Ethernet is used in more than half of all LANs installed to date. On an Ethernet network, the user's PC "listens" to the line or highway before communicating. when the line is quiet or free, the message is transmitted so it won't collide with another message. Like Token Ring, this happens relatively quickly. The speed limit of Ethernet is 10 Mbps. An Ethernet highway is paved with coaxial or UTP cable and can support over 1,000 drivers or devices, given certain length restraints.

 

Local Talk

Most commonly associated with the Apple environment, LocalTalk is built into every Macintosh computer. As with Ethernet, the user's PC "listens" to the line or highway before communicating. When the line is quiet or free, the message is transmitted so it won t collide with another message. The highway for LocalTalk is paved with UTP cable and its speed limit is 230 kilobits per second (Kbps). Under the right conditions, a maximum of 48 drivers or devices can reside on a LocalTalk highway.

Network Cable Characteristics

Network Backbone Architecture

The network architecture is the supporting structure, or infrastructure, that basically supports network computing. It comprises several major subsystems, including the layout or topology of the network, cabling, and intranetworking devices, such as bridges, routers, and switches. When designing a network, each of these network resources must be considered to determine how to implement and distribute them in a way that optimizes performance, facilitates management, and accommodates future growth.⁷

The cabling system is among the most critical set of network components. A poorly designed or installed cabling system can be difficult to troubleshoot and can cause many needless hours of downtime. A good cabling system, on the other hand, is designed for reliability, manageability, and maintainability. Older, linear cabling systems, such as the original Ethernet and thin Ethernet systems, were not designed with troubleshooting in mind. In contrast, modern, star-wired systems are much easier to manage and maintain. (Corrigan)

The ability of an expensive, high-end network environment to provide reliable communications can be notably compromised by insufficient cable wiring. Analysts speculate that at least 50% of networking problems are the result of faulty and improper wiring decisions, and corporate IT managers must ensure that wiring receives the same level of technological support as other components of a network. Network managers should consider their companies' present infrastructure, plan for future expansion, carefully detail wiring configuration plans, conduct thorough tests and implement wiring management devices. ⁸

Several network backbone alternatives have come to the forefront as leading technologies for current and future backbone alternatives. They include Fast Ethernet, (Asynchronous Transfer Mode) ATM, and Gigabit Ethernet.

Fast Ethernet is the most common choice at present for LAN backbones to the desktop, in part because most networked companies are familiar with 10-Mbps Ethernet.⁹ Fast Ethernet speeds up the network by a factor of ten, while simultaneously decreasing the diameter of the collision domain by ten. All major networking vendors now support Fast Ethernet, although LAN designers must still take into account future needs and limitations when designing the Fast Ethernet LAN. ¹⁰

100Base-TX twisted-pair cabling must be installed by CAT-5 certified cabling installers and tested by 100MHz cable testers that adhere to the EIA/TIA TSB-67 guidelines. The patch cables at both the work area and the wiring closet should be included in the testing. The primary CAT-5 cable tests are attenuation and crosstalk. Crosstalk should be tested from both ends of the horizontal cable run. Any deviation on cable type, installation, or testing will have serious negative effects on 100Base-TX Fast Ethernet. (Cali)

 

Fast Ethernet switches can provide a limited amount of raw bandwidth, but ATM backbones have the potential to save networks inundated with multimedia demands. ATM expands high-bandwidth lines and manages multimedia by dividing data into small, fixed-size cells that make traffic less congested and more predictable. While Fast Ethernet is less expensive and easier to integrated into existing networks, ATM is a competitive solution when creating a network from scratch. Additionally, the ATM Forum is developing standards, specifically the Multi-Protocol Over ATM standard, to help simplify the process of running Ethernet over ATM switches. ATM Forum members are also seeking ways to make IP connections run more smoothly with ATM technology. Despite these advances, ATM's benefits cannot be fully utilized until there is software, which requires a solid API. ¹¹

The Gigabit Ethernet Alliance vendor group has finalized an open, cost-effective standard for 1Gbps Ethernet that will be known as IEEE 802.3z. The specification should allow for dramatic growth of the new technology, which leverages the market dominance of standard Ethernet and is attractive to corporations because it lets them upgrade to high-speed backbones without the costs associated with ATM migration. Standardization will let Gigabit Ethernet further extend the scalability of LAN backbones and connections between servers and switches. Many analysts view it as complementing rather than competing with ATM, which is used more on WANs than LANs. It will be deployed first on the backbone. IEEE 802.3z will allow half- and full-duplex operation at 1,000Mbps while relying on the existing Ethernet CSMA/CD access method and frame format. A second draft will ensure support for unshielded twisted pair (UTP) wiring. ¹²

Standard Wiring Diagrams

Wiring diagrams for a typical school classroom, campus computer lab, network wiring closet (IDF), and wide-area network are included in Appendix B.

1. Network and Communication System Integration

Premise: Existing computer systems should be integrated as much as possible into the communication system of the district.

Integrating Voice and Data

From a technical point of view, integrating voice and data is not an issue. The issue is whether the requirements exist to do so. Before you decide that integration is where you need to go, have an expert evaluate your current voice system(s) and make recommendations. GTE and US West are two companies that offer this service on a no fee basis. According to service providers US WEST and GTE, a rule of thumb for consolidating multiple OPX (Off Premise Exchange ) circuits into one T1 circuit is when the number of lines reaches or exceeds 6. A T1 circuit has the ability to carry 24 circuits. The reason the number 6 is used is that one OPX circuit runs in the neighborhood of $40-$50 per month. T1 circuit costs are approximately $250 per month. For the price of one T1, you gain 18 circuits. The amount you pay is dependent on many things; where you are located, who your service provider is, distance from your site to the providers CO (Central Office) and other factors. The number of circuits dedicated to voice and data are controlled by the user. ¹³

The increased interest in integrated voice-data (IVD) is due in large part to the belief among many industry analysts that the technology will become an important aspect of future network upgrades. Others claim that IVD will cost too much and lack the quality users require. IVD treats voice as just another client/server application on a data network, but the technology differs from computer-telephony integration (CTI) in that it is said to reduce infrastructure costs by eliminating duplication of wiring and by replacing proprietary PBX hardware with software. Other cost savings come from bypassing the local loop and diverting expensive voice traffic to unused data capacity. IVD may require that sound cards, microphones and speakers be added to PCs, but this will be made easier by Intel's support for the Universal Serial Bus, and by the WinSock 2 API. ¹⁴

Attempts to integrate voice and data lead to conflicts between MIS and LAN managers because IS seeks the cheapest solution for handling voice calls while those on the LAN side want to use the best practices available. Internet-based solutions may ultimately predominate because they use standardized technology and are flexible. MIS managers generally assume that new voice/data services cost the same as conventional phone services and have the same reliability. Voice-over-Internet vendors have real, viable products on the market, while frame relay is at last becoming interoperable. Internet schemes, unlike frame relay, do not dictate a solution because they can run over many types of physical and transport networks. Cost and reliability are interrelated more than MIS managers tend to realize. Making PC-based voice systems as robust as conventional PBXes will be expensive, requiring considerable coding and integration. ¹⁵

As demand increases for high-speed networks capable of managing both data and voice, the telecommunications network will be transformed at a rapid pace. Telecom-service providers, such as the RBOCs, are forced to use multi vendor equipment to reduce their time to market. The International Telecommunications Union (ITU), in response to this systems-management dilemma, has defined the telecommunications-management network (TMN) specification to standardize management practices worldwide. TMN standards specify five hierarchical management layers, which include service, business, network elements, network-element management and network. TMN also defines two roles, agent and manager, inside the management structure. Agents communicate with managers via the Q3 interface, which is made up of the common management-information protocol operating on top of an open systems interconnection stack. The advantages of a TMN solution include decreased deployment time, task simplification and increased network reliability. ¹⁶

The ability to transmit voice over LANs may ultimately make the traditional PBX obsolete and absorb PBXs into computer systems. Fixed latency in a non-blocking environment is the key enabling technology for voiceLAN because it ensures that voice messages are delivered in real time. User acceptance is as essential as physical architecture in determining the success of the new technology. There will be more off-the-shelf voiceLAN products and network and phone-ready PCs available as the technology advances. It is up to leading-edge customers to fuel demand, something that will not happen until clear advantages of integrating voice into shared applications are apparent. Adding voice capability to white boarding is a long-promised benefit, as is full integration of E-mail and voice mail. Cost may seem high at first, especially since most businesses already have a PBX installed, but shared applications and multimedia already demand hardware changes and installed voice wiring will soon need replacement due to age. ¹⁷

2. Network Server Centralization

Premise: Network servers should be located where they can be easily and remotely managed and supported.

"Is it better to centralize or decentralize your networking devices?" Due to recent trends the pendulum is swinging again, this time back to centralization.

* Fiber-optic backbones. Either as a vertical riser or a campus backbone, fiber is clearly the superior choice over Cat 5. Today, the cost of installing fiber is just marginally more than Cat 5, and yet it provides far superior security, distance and overall future-proofing. Now that the cost and ease of installing fiber is so close to that of Cat 5, more organizations will be deploying it.

An organization with a fiber backbone has the opportunity to overcome the greatest challenges to a centralized network architecture: local performance and backbone congestion. With fiber, switching traffic between LAN segments on a single floor via a centrally located switch will occur at such a high-performance level that it's imperceptible to end users. As far as they are concerned, the switching appears local, even though it's not. In addition, the capacity of the fiber backbone is so great that most organizations can use it for access to a centralized switch without being concerned about total bandwidth. As organizations install fiber in the vertical or campus, many will reevaluate their preferences for a centralized or decentralized network.

* Big switches. The preponderance of LAN switches available to date have been 16 ports or fewer. Most have had, at best, one or two high-speed uplink ports. These products aren't appropriate for central locations, so early LAN switch adopters didn't have a lot of choices regarding network architecture. Today, a wide variety of vendors offer LAN switches that provide support for multiple high-speed uplinks. Switches are finally becoming available that have the high-speed port density and backplane capacity needed for a centralized network.

* 100-Mbps LANs. Any organization that sees 100-Mbps LANs in its future, whether it's a few or many, will need to consider even higher-speed LANs for backbones or access to centralized resources. Regardless of whether it goes with ATM or Gigabit Ethernet, an organization with such a high-speed network will need specialized products.

Because of the cost and management needs of these devices, they also favor a centralized approach. So organizations that are thinking ahead and planning for 100-Mbps and even higher-speed LANs also need to consider the implications for the backbones that will support them. This, too, points towards a centralized network approach. ¹⁸

The following several pages are borrowed from a white paper from Synoptics Corp. Entitled "Switching Paradigms - Everything Has Changed Except the Network".

 

Today's broadcast-based shared media networks are buckling under the load of new applications and the demands of networked organizations. A new paradigm in networking-high-speed frame and ATM cell switching-promises to meet the ever-increasing demands for network services. Switching represents a significant evolutionary step beyond traditional LAN internetworking and will provide the performance, scalability and ease of management required by the modern information enterprise. This document explores the forces driving users to adopt switching technology and explores many of its benefits

Everything Has Changed Except the Network

Business use of information technology has undergone a radical transformation in the last decade, from back room to customer facing; from cost-counting to revenue producing; from infrastructure tool to agent of change; from luxury to competitive necessity. As a result, almost every aspect of corporate IT has changed -- tools, technology, skills, methodologies, practices and plans. Every organization, it seems, is re-engineering its business functions by downsizing, migrating to client/ server architectures, using advanced graphics, or exploring imaging technology. Inspired by business needs, such IT initiatives make organizations increasingly dependent upon their networks. Although networks are but a single element in the IT environment, they are, perhaps, the most critical, serving as the corporate nervous system through which command, control and communications are exercised.

Networks Losing Ground

Only an industrial-strength, production-quality network is an acceptable platform for supporting mission-critical applications tied to a revenue stream that is central to an organization's competitiveness. Unfortunately, most networks do not provide a consistent level of high-quality service. Superior performance, scalability and manageability are noticeably absent.

The reason is simple. Although the IT environment has undergone a radical transformation, networks have evolved little. With the arrival of client/server computing, the basic application architecture has changed. Groupware represents an entirely new paradigm of application usage. Network traffic patterns are shifting dramatically, and processing power, as well as the number of connected users, is growing rapidly.

While not every reader will recognize such trends in their organization, leading technology adopters have discovered that their networks are in trouble right now. Standing at the edge and looking back, they see that there is no quick-fix for these problems. Only a comprehensive networking paradigm shift can get them back out in front of the demand curve again.

Client/server computing is rapidly becoming the mainstream application architecture. The increase in flexibility and reduced costs associated with the client/ server architecture makes it a win for the IT organization and the business. Unfortunately, an unprepared network may be the undoing of applications planning. Emerging applications require that networks move more data in less time, across larger geographical areas and under heavier transaction volumes than ever before. Today's networks are largely incapable of meeting such demands while providing the reliability required for supporting these mission-critical applications.

Subroutine Calls Become Network Transactions

Client/server computing breaks the mainframe monolith, under which the network and the host system were virtually indistinguishable from one another. Under the client/server architecture, what once were subroutine mainframe application calls (executed entirely on the host system) are now client/server transactions performed across the network. This process makes these applications much more dependent upon the network.

 

Downsize the Computing; Upsize the Network

Many organizations, seduced by the promise of considerable cost savings, have migrated their applications from mainframes to client/server environments. However, users and IT managers alike would be well advised to consider the following case before counting these savings.

Figure 1

Illustrates a task in the terminal/host and client/server environments. A40-byte user query receives a 1,000 byte response. Unfortunately, due to protocols and distributed processing, the client/server application needs 10 times the bandwidth required in the terminal/host environment. Bottom line: don't move your mission-critical applications to client/server networks without investing in your network.

Terminal/Host: Shaded area "A", in Figure 1 at left, indicates elapsed time for the 40 byte request to travel from terminal to host. Shaded area "B" indicates elapsed time for host processing, and "C" represents the transmission of the l000-byte response. To complete this transaction in two seconds, the terminal/host network requires roughly 10 Kbps bandwidth.

Client/Server: The 40-byte client/server request ("D") is now followed by a server/client acknowledgment ("E")--traffic the terminal/host network did not need to carry. After sending the acknowledgment, the server formulates the l,000-byte response ("F" = processing time) and prepares to transmit to the client. However, the protocol used in this example requires the server to break the 1,000-byte response into two pieces, each sent independently over the network and followed by an acknowledgment of receipt. Dividing the processing across two machines takes data transmission time away from the network, and sending acknowledgments means the client/server system must move more data in less time than the terminal/host system.

The Difference: In order to complete the same application task, the client/server network requires 108 Kbps of bandwidth--ten times more than the terminal/host. The bottom line is that if you downsize the computing, plan on upsizing the network! Dollars saved on host processing should be reinvested in the network infrastructure if your applications are to provide response times that are consistent with user expectations.

More Data, Less Time

Regardless of increased network dependency, neither users nor developers are willing to sacrifice any additional time for the applications and network to complete their tasks. A certain bravado exists among developers about their applications performance, while users have grown to expect the rapid response times they enjoy in mainframe environments. Even enthusiastic users who are sold on the flexibility of client/server computing are likely to be disappointed if they perceive the system as being slow. That means that, almost by default, client/server systems must move more data in less time than host-based systems in order to be considered "successful".

Application/Network Design and Transaction Loads

Modern applications must be designed to run on networks. Network traffic is generated as users perform various tasks such as browsing entries, painting screens, or even moving a mouse. Good applications can be as much as 10 times more efficient in using network resources than sub-optimal designs. A further consideration is that as client/server applications become more powerful, application procedures and data are distributed across larger numbers of clients and servers. Increases in distribution result in even greater reliance on the network. In such environments, a single user query to a local server often triggers many server/server transactions. Add to this the increasing complexity of ad hoc queries generated by experienced users and suddenly the network is required to support cascading sets of transactions just to respond to a single inquiry. As a result, client/server networks must often cope with many more transactions per user than mainframes do.

Server Centralization Adds To Network Latency

As client/server networks begin to support mission-critical applications, issues such as security, reliability, disaster recovery and control become increasingly important. In a client/server environment, servers require the same type of protection afforded mainframes and minicomputers. Normally, this is accomplished by installing the servers in centrally-located, environmentally-controlled rooms complete with halogen fire protection and redundant back-up power supplies-the same place where the mainframe was once located.

Centralizing servers moves them from the distributed workgroup LANs onto data center (server center) building or campus network backbones. Routers and bridges are the line of demarcation between the backbone and the LAN, which means that centralizing servers force all client/server transactions to cross these devices. Unfortunately, the vast majority of today's bridges and routers are not designed to handle the huge traffic loads generated by production client/server applications. Although some bridges and routers meet the gross performance requirements of moving lots of packets, crossing a bridge or router en route from client to server adds considerable latency to the process, slowing the application and hurting user response time. In response to this problem, many network managers demand bigger, faster routers. Unfortunately, these managers don't realize that bigger, faster bridges and routers are only a temporary-and expensive - solution.

The Details: Client/Server Distance And Network Design

A critical network/application design consideration is the distance between clients and their associated servers. As a rule, the greater the distance, the greater the bandwidth required to complete a transaction in a fixed time interval. This rule applies equally to both building and campus networks and to the WAN. Figure 2 at right illustrates the time it takes to complete the same client/server transaction in three different scenarios.

Workgroup Client/Server: Completing the benchmark transaction along path Number 1 requires 1.1 seconds.

Centralized Server: Completing the benchmark transaction along path Number 2 (a server centralized in a data center) requires 1.6 seconds--a 50 percent increase in network response time. Bridges and routers interconnecting shared LAN segments and backbones introduce delays which degrade application response time. Adding back server processing time and paying the additional network delay across several transactions compounds this problem.

Figure 2

A "typical" client/server transaction shown in three different network environments illustrates the effect of separating clients and servers across a network. More distance means longer transmission delays and degraded response times. Network/application designers must often balance the cost and reliability issues of replicating servers against the speed and performance of the network.

Remote Server: The WAN scenario makes the point in the extreme. Because of greater distances and lower line speeds, the campus/building problem is exacerbated, and completing the benchmark transaction along path Number 3 now requires more than eight seconds.

The differences in response time are important application/network design issues for a number of reasons. Application developers want a single copy of the data to exist on a centralized server to ease database administration. Unfortunately, significant latency is introduced when queries are routed back to a single server. There are only two methods for alleviating this problem: (1) Redesign the application and put data on multiple distributed servers near large concentrations of users; or (2) increase the speed of the network to compensate for client/server distances. The best way to get a handle on this problem is to validate application design using realistic network testbeds.

 

 

Bottlenecks At Critical Resources

The centralization problem is, of course, not unique to client/server computing. Any critical network resource may become a bottleneck as more and more users compete for the right to connect. The most logical way to relieve these congestion points is to increase the carrying capacity of the individual links to such devices. Historically, however, network planners have not been able accomplish this with Ethernet and Token Ring. Instead, installing a high-speed server connection required replacing the server's Ethernet or Token Ring interface with an FDDI card and then either (1) giving all connected workstations an FDDI interface or (2) using a router to move traffic from Ethernet- or Token Ring-connected users to the FDDI-connected server. Neither solution is simple nor cheap; the network manager's only option is to choose between buying a new high-performance LAN or buying more internetworking technology.¹⁹

Whether or not an organization supports mission-critical client/server systems, it is certainly subject to a few universal IT trends. One of the most significant is the growth in the number of connected users. Compounding the growth problem is the fact that most new users get access to the newest applications (often client/server), further complicating the problem of a major network build-out. A parallel trend is that traffic volume, even across a fixed number of users, increases over time, further stressing networks. Experienced users push their applications further, send more e-mail messages, and invent more complex queries as they adapt to life on-line-the inescapable result of effective user training. (Serjak)

Just as important as user population growth is the increase in processor performance. Moore's Law states that microprocessor density doubles roughly every 18 months. This equates to improving price/performance by a factor of 2 in the same interval. Newer, faster machines have 1,0 capabilities that are leveraged in servers and high-performance workstations. Chips like Pentium, Alpha and SuperSPARC can easily drive traditional shared-media network connections to the point of saturation. High-performance servers need high-performance networks if anyone expects to derive benefits from them.

A final consideration for the network planner is the growth in size and types of data used by modem applications. The use of multimedia, video, voice and data will challenge networks to handle different classes of data according to the specific needs of that class. Voice and video are highly sensitive to network delay, while data is not. While desktop video is enticing, it's often hard to make a solid business case for deploying it. imaging, however, is another story. Companies ranging from banks to insurance companies, hospitals to aircraft manufacturers, are rapidly migrating to electronic storage and retrieval of documents because it meets a primary business need: it saves money.

In the final analysis, networks are falling behind. Advances in the number of users, average bandwidth per user, processing power and evolutionary changes in application data types are outpacing existing networks' abilities to support them. (Serjak)

Repealed: The 80/20 Rule of Network Design

Another problem facing most corporate networks is that modem applications and models of application/network usage break one of the few historically reliable laws of networking: the 80/20 rule. The 80/20 rule states that 80 percent of all network traffic will remain in the local area while 20 percent is destined for a remote location. While this rule has historically held for both voice and data transmissions, the business goal of creating smaller, decentralized, flexible, empowered teams repeals it. Employees must have access to network resources, regardless of their-or the resource's-location. The any-to-any connectivity paradigm has been recast as the everywhere-from-anywhere networking paradigm. Network designers no longer have the luxury of assuming that users will be connecting to local resources. This means that the center backbone) of the corporate network must have much more capacity than ever before in order to move traffic high speeds over long distances between user and resource. In the extreme case, backbone capacity must approach the sum of the capacity of all workgroups in the enterprise. Unfortunately, corporate backbones have not been built with this assumption in mind. (Serjak)

Universal Goal: Scalability and Simplicity

Scalability is one of the principal goals of corporate networks, along with manage ability and afford ability. Scalability is a measure of network resiliency in the face of increased demand. For a network to be considered scalable, the principles that govern its operation and growth must remain in effect as the size or scope of the network changes. However, increasing complexity means today's networks become more difficult to administer as they grow in size. As with other types of systems, simple networks scale better than complex networks.

Corporate networks are comprised of diverse networking technologies combined in such a way that they appear, to the user, to interoperate seamlessly. Five years of grassroots PC LAN deployment, combined with legacy systems, have created a diverse mix of technologies, protocols and standards. This mixture is often little more than the juxtaposition of one system on another, resulting in a network that is difficult to manage, change and grow. Today's enterprise networks have not been constructed, they have been crafted. Each change in their structure challenges the skill of the network craftsman to reshape the system into a useful tool. Because the requirements driving change never cease, the network is never quite complete, and a certain frailty remains. Frailty and complexity are diametrically opposed to simplicity and scalability. For more information, see the sidebar entitled The Details: Microsegmentation and Scalability (Serjak)

Same Old Networks?

Everything has changed except the network. Downsizing, the adoption of client/server computing, centralized resources, new classes of applications, new usage paradigms, more connected users, and more powerful computing platforms have all contributed to the new IT environment. This new environment seeks more from today's networks than they can give. Leading adopters have done all they can to squeeze as much performance as possible out of existing networks, but it has become clear that incremental fixes will not solve the problems listed above. Existing shared media networks, based solely on bridges and routers, have run out of gas, creating the need for a new networking paradigm. The answer, it seems, is to build networks based on switching technology.

The Details: Microsegmentation And Scalability

Increasing network bandwidth is a major goal of network scalability. Unfortunately, there is no graceful way to accomplish this with existing networks. Traditional Ethernet and Token Ring networks are broadcast networks in which all end-stations on the LAN receive all frames transmitted and discard those that aren't specifically addressed to them.

Bridges and routers can filter traffic received on a shared LAN interface and forward frames to other LANs only when necessary. Therefore, the conventional method for dealing with an oversubscribed LAN is to split it in half and connect the two segments with a filtering bridge or router. The traffic loads on each of the new segments is then smaller than it was on the single large segment.

Enterprise networks with large bridge- and router-based internetworks have repeatedly split their LANs into smaller and smaller pieces--an approach called microsegmentation--to increase available capacity and improve performance. While a major site might have used a single Ethernet to connect 300 users five years ago, a typical segment today supports around 20 end-stations, and is quickly headed toward eight to 10 end devices.

(See Figure 3.)

Consider a 100-node Ethernet LAN. Since subscribers take turns using the network, any individual end-station is only guaranteed 1/100th of the total available 10 Mbps of the Ethernet bandwidth, or about 10 Kbps. If the Ethernet is saturated, the network may be divided in two with a bridge or router, creating two segments with 50 users each. Now, each user receives 1/50th of the Ethernet bandwidth, or roughly 20 Kbps. As the network is chopped into smaller and smaller pieces, each containing fewer users than the original, individual users get proportionately more bandwidth.

Unfortunately, microsegmenting networks with routers quickly becomes a game of diminishing returns. Microsegmentation creates a management headache because the number of routers and network segments increases. Segmentation probably requires changing end-station addresses each time a new segment/subnet is formed. And even FDDI falls victim to segmentation, since it is a shared-media LAN solution, just like Ethernet. There is a limit to the number of times a LAN or backbone can be divided. In the end, each network device gets its own private router or bridge port.

The way to increase capacity without introducing complexity is with switching

Figure 3

Microsegmentation at Work. To get more capacity out of a shared-media LAN, managers divide it into smaller segments, each containing fewer end station. The increasing number of segments and routers leads to complexity, making management more difficult and thwarting scalability. managers need a way to increase capacity without chopping the network into pieces with routers. Switching seems to be the answer.

 

 

The Details: Frame Switching

A frame switch operates much like a multiport bridge. Switches learn the address(es) of each attached end-station. By examining the destination MAC address of each packet it receives, the switch can then forward the packet to the output port attached to the end-station with the same MAC address. The output port may be attached directly to the end-station, to a shared media segment with multiple users, or the port may be connected to another frame switch which, in turn, is connected to the destination. There are, however, a few notable exceptions.

While bridges and routers are best at interconnectin