Ian & Stuart's Australian Mac: Not for Sale

home *** CD-ROM | disk | FTP | other *** search

/ Ian & Stuart's Australian Mac: Not for Sale / Another.not.for.sale (Australia).iso / fade into you / being there / Rants / Understanding Switching < prev

Wrap

Text File | 1994-09-18 | 84KB | 1,420 lines

Understanding Switching and Virtual Networking THE BANDWIDTH CHALLENGE Not long ago, a single segment supported all users on the corporate LAN. The typical user device was either an asynchronous terminal or an early generation PC or workstation. In this environment, where few applications other than terminal emulation and network printing were used, a single LAN segment or ring provided sufficient bandwidth for all users. Today, network traffic continues to increase, not only because of the large number of users but also because of the wide range of high-traffic applications supported. Network administrators are being pressured to provide increasing bandwidth for users while keeping an eye on evolving technologies such as ATM, preserving as much of the existing infrastructure as possible, and living within budgetary constraints. Wiring hubs that incorporate multiple LAN networks in combination with internal or external bridge/routers are some of the solutions being used to increase effective bandwidth and network manageability. Other technologies are also playing an important role in increasing network throughput - some of these are already developed and mature and some are currently emerging. These solutions include a range of technologies such as FDDI, module switching, port switching, LAN switching, and ATM switching. Virtual networking is also emerging as a tool to improve network manageability. THE SOLUTION SPECTRUM With a number of different approaches available for improving network throughput, a network administrator must weigh the merits and cost of each solution. Many networks contain almost the full spectrum of available technologies, with the higher performance and higher-cost solutions existing at the top of the network pyramid (see Figure 1). xref image = fig1.tif Figure 1: The Connections Pyramid Understanding Switching and Virtual Networking It is rare for a large network to require the higher-speed, costly technologies such as ATM and FDDI for each and every client station. It is rarer still to find an organization that has a virtually unlimited budget allowing it to incorporate the latest technology throughout the network infrastructure. Increasingly the challenge for network administrators is to identify the best technology for supporting a new or emerging application and to blend this technology with the existing network. Low-speed Shared LANs Traditional LANS, such as Ethernet and Token Ring, can be found in almost every network and will continue to play a major role in future networks. Network congestion occurs in these low-speed shared LANs when a single segment or ring is no longer able to support the generated traffic in a timely fashion. The ability to segment a LAN into smaller, more manageable workgroups and to limit the traffic between these workgroups are critical components in improving network throughput. LAN segmenting and traffic filtering are accomplished using bridges and routers (see Figure 2). Due to their ability to create multiple segments or rings, bridges and routers can reduce network congestion by effectively increasing the total amount of bandwidth available to each workgroup. xref image = fig2.tif Figure 2: Low-Speed Shared LANs Wiring hubs, in combination with network management, also play a role in reducing network congestion in low-speed shared LANS. Wiring hubs support growing user populations in a structured fashion, similar to the star-wired telephone system, and provide improved problem diagnosis and resolution. Each port on a wiring hub can be monitored for traffic loads and packet characteristics using network management tools, allowing a network administrator to pinpoint heavy traffic producers and to develop physical groupings around traffic patterns. When to use low-speed shared LANs Low-speed shared LANs combined with internetworking devices provide an effective solution for reducing network congestion in small- to medium-sized networks. As networks become larger and/or traffic increases, the number of available segments or rings becomes a limiting factor - both in physical connections and in the amount of bandwidth available to support traffic. Other solutions, often in combination with low-speed LANS, need to be considered to address congestion issues. Understanding Switching and Virtual Networking High-speed Shared LANs One alternative for reducing congestion on a network is simply providing each user with higher-bandwidth connections. Network technologies that provide high-speed user connections include FDDI, TP-DDI, and two proposed standards for 100 Mbps Ethernet - 100Base-T and 100Base-VG. FDDI offers mature standards and products that are available from numerous vendors. FDDI is typically used for creating high-speed backbones to connect lower-speed technologies and for direct connection of servers and power users. It is used to a lesser extent for client connections due to the cost of installing fiber optic wiring and equipping workstations and PCs with new adapter cards. Per port cost for an FDDI connection is approximately $2000. The recently completed TP-DDI standard for operating FDDI protocols over unshielded twisted pair (UTP) wiring may spur the use of FDDI protocols for client delivery. However, even though TP-DDI may allow existing wiring to be used, the commonly employed 10 Mbps Ethernet and 4/16 Mbps Token Ring network interface cards (NICS) and transceivers would still need to be replaced at each end station. Today, it could cost between $1500 and $1700 per port for a TP-DDI connection. The emerging 100 Mbps Ethernet standards also promise to provide high-speed shared media networks for end-user applications. But although the IEEE 802.3 10OBase-T and IEEE 802.12 10OBase-VG proposals maintain Ethernet ftame compatibility, they require new NIC cards at each end station and will cost approximately $ 1000 per port to implement. While both proposals intend to support operation over Category 3 and Category 5 wiring, only the IEEE 802.12 proposal currently demonstrates Category 3 operation. The widespread commercial deployment of 100Base-T and 100Base-VG shared LANs will not occur until devices such as concentrators, switches, and high-end bridge/router interfaces, with proven interoperability, are available from multiple vendors. When to use high-speed shared LANs High-speed shared LANs are most appropriate for supporting applications where multiple clients require high-speed access to servers and/or to each other, particularly under high burst rate conditions. In this type of application, lower-speed shared media can support users requiring speeds of 10 to 16 Mbps, while the high-speed shared media can support users requiring burst speeds in the range of 100 Mbps. High-speed shared LANs are most commonly used to address the specialized requirements of small, high-speed client/server networks. In most corporations, these networks are usually limited to high-end engineering, medical imaging, and graphics-oriented applications. While high-speed shared LANs do increase the available network bandwidth, they are expensive to implement for every station on a network. The successful use of high-speed shared LANs assumes that adequate wiring, typically Category 5 twisted pair or fiber optic cabling, is already installed or can be installed. High-speed network interface cards are also required for each client station and server to be connected to the high-speed network. A more Understanding Switching and Virtual Networking cost-effective solution is to divide a single LAN into multiple segments and use internetworking devices in combination with high-speed backbones to provide communication among all devices. High-Speed Shared Backbones High-speed backbones can be used to interconnect multiple lower-speed hubs, each of which supports groups of clients over low-speed shared-media interfaces, such as 10 Mbps Ethernet (see Figure 3). Servers can also be connected to the hubs or directly attached to the high- speed backbone. While this approach does not provide higher bandwidth to individual client stations, it can help eliminate throughput bottlenecks to shared computing resources, such as servers, as well as provide a high-capacity channel to service the communications requirements of a large number of hubs. FDDI is often used in this configuration, providing a 100 Mbps shared communication backbone. When operating over fiber, FDDI is capable of spanning large distances and providing a high level of fault tolerance. This configuration is referred to as a distributed backbone network. xref image = fig3.tif Figure 3: FDDI as a Distributed Backbone Alternatively, FDDI (or TP-DDI) can be used in a local or collapsed backbone configuration - potentially making use of a single FDDI (or TP-DDI) concentrator to implement the high- speed backbone (see Figure 4). Some wiring hubs also include an internal FDDI ring. Understanding Switching and Virtual Networking xref image = fig4.tif Figure 4: FDDI as a Collapsed Backbone When to use a high-speed shared backbone High-speed backbones are appropriate when multiple hubs located in a single building or campus environment need access to high-speed server(s) and/or each other. A primary advantage to the high-speed backbone approach is that the wiring and NIC cards at the end stations can be maintained. To access the high-speed backbone, one high-speed interface is required per department or workgroup hub - typically in the form of a bridge or bridge/router - to convert the traffic on the lower-speed network to the appropriate format used on the high-speed backbone. High-speed backbones based on FDDI have been widely deployed and most vendors of networking and internetworking equipment support FDDI interfaces. FDDI is a mature technology with proven interoperability and declining connection costs. There are large FDDI networks that efficiently connect many thousands of network users. For customers concerned about future bandwidth availability, a star-wired fiber optic configuration allows a smooth migration to future LAN switching or ATM switching technologies. Fiber optic cabling is also an attractive choice in facilities where conduit space is scarce. Collapsed Backbone Bridge/Routers Another popular form of collapsed backbone uses high-performance multiport bridge/routers. Using high-end bridge/routers allows individual segments or rings to be interconnected with Level 3 routing as well as Level 2 bridging when required. Routing between segments and rings allows network administrators to control bandwidth use and to enforce access and security provisions. It also allows network administrators to implement mixed media networks, such as networks supporting both Ethernet and Token Ring. Understanding Switching and Virtual Networking Many collapsed backbone bridge/routers provide the ability to connect to high-speed backbone networks, typically an FDDI backbone. Collapsed backbone bridge/routers also support a variety of WAN interfaces, allowing them to collapse both LAN segments and distributed WAN networks. Because collapsed backbone bridge/routers are designed to be all-purpose, the cost per port Is relatively high. Typically, high-end bridge/routers are designed so that each port is capable of supporting a full segment or ring of bridged or routed traffic. Each interface usually supports from four to twenty routing protocols and provides a wide array of Level 3 traffic and security control features. When to use collapsed backbone bridge/routers Collapsed backbone bridge/routers are very good choices for connecting multiple medium- to high-traffic network segments together and enforcing access and security rules between stations. Because of their ability to support mixed media and WAN interfaces, collapsed backbone bridge/routers are useful in large enterprise networks. Because of their high cost per port, however, current collapsed backbone bridge/routers are not cost-effective choices for microsegmenting LANS. A better solution for improving network throughput is to combine various switching technologies for segmenting networks at the client end with a high-speed backbone or collapsed backbone bridge/router. Switching Technologies Dividing a single LAN into multiple segments and using internetworking functions to tie these segments together increases the available bandwidth to clients and servers. Historically, network segmentation was achieved using physically separate units, such as multiple wiring hubs connected via two-port bridges or bridge/routers. This implementation is now being replaced with high-performance, multipart collapsed backbone bridge/routers. There is also a growing trend toward providing multiple segments in a single chassis and tightly integrating bridge/router and/or switching functions into the enclosure to provide a much more cost-effective implementation for improving network throughput. A number of different architectures have emerged that offer differing levels of bandwidth improvement, flexibility, and cost-effectiveness. Despite their fundamental differences, these products are all referred to as switches. Switching Defined The term switching means different things to different vendors - creating confusion in the minds of network administrators. Generally, all switching approaches make use of some type of parallel communication. One way the term switching is used is to describe the ability to connect (switch) users, on a module basis or on an individual port basis, to one of the multiple LAN segments or rings within a wiring hub. This type of switching is typically performed by a network administrator during configuration and is only modified if, for example, a user moves or there is a fault on a LAN segment. This technology can be viewed as static switching. Module switching and port switching, described in this paper, fall into this category. Understanding Switching and Virtual Networking Another way the term switching is used is to describe the ability to connect a transmitting station to a destination station "on the fly", or in real time. This type of switching allows multiple device pairs to communicate simultaneously through a high-speed switching fabric. This switching approach can be referred to as dynamic since connections are established on an as-needed basis. LAN switching and ATM switching fall into this category. Any of these switching technologies, used alone or in combination, can play a part in reducing network congestion. Module Switching Module switching describes the ability to connect all the devices attached to a given module to one of the physical LANs within a single hub (see Figure 5). Current generation wiring hubs typically provide between two and five internal segments or rings, providing two to five times the bandwidth of a single LAN. For example, if a hub contains three internal LANS, a network administrator could provide groups such as marketing, finance, and engineering with their own LANS. The users from each department would be connected to one or more modules which are then connected (switched) to a designated LAN. Module switching provides a straightforward way to distribute users across multiple LANs within a wiring hub. A major benefit of 10 Mbps Ethernet and 4/16 Mbps Token Ring switchable modules is that they support the existing wiring, adapter cards, and end-station software. One limitation is that all users within a department or workgroup must be located within the physical wiring distance of the hub. xref image = fig5.tif Figure 5: Module Switching Understanding Switching and Virtual Networking In module switching, devices connected to the ports on a concentrator or MAU module may be switched among the wiring hub's internal LANs via network management control. Using this approach, all devices connected to a concentrator or MAU module must switch to the same LAN at the same time. A bridging and/or routing function is needed to connect these internal LANS, providing interoperability among all devices on the network. While the internetworking function may be provided by an external device, some wiring hubs support both module switching and integrated bridging and routing. A single integrated device provides a more cost-effective and easier-to-manage solution for creating multi-segment or multi-ring networks. When to use module switching Module switching used in combination with integrated bridging and routing is ideal for upgrading single-LAN networks that are experiencing throughput problems. Dividing a single-LAN network into multiple workgroups and connecting them with a bridge or router will improve each workgroup's throughput. Module switching is effective when each workgroup has its own file server so that constant access to a single file server is eliminated. It is also effective when the internetworking device connecting the internal segments provides either multiple connections to shared resources or a high-speed connection to shared resources (see Figure 6). xref image = fig6.tif Figure 6: High-Speed Connection to Shared Resources Understanding Switching and Virtual Networking Port Switching Port switching uses multiple physical LAN segments or rings in a single intelligent hub (see Figure 7) but offers more flexibility than the module switching alternative. Port switching allows the network administrator to switch users on a per port basis, not a per module basis, among the internal LANS. xref image = fig7.tif Figure 7: Port Switching While the capital investment is higher for port switching than it is for module switching, port switching reduces labor costs by simplifying adds, moves, and changes. Individual users can be moved among the available internal LANs via network management control, not by making physical changes in the wiring closet. For example, if an office that was occupied by an accountant is now occupied by an engineer and module switching is used, the network administrator would have to physically move connections in the wiring closet to implement the change. Using port switching allows a network administrator to be more responsive to users' requests for connectivity changes and is also useful for fault isolation on a per-port basis. Like module switching, port switching is compatible with existing wiring, adapter cards, end-station software, and backbone structures. When to use port switching Port switching, like module switching, is effective for reducing network congestion in single-LAN networks. In addition to allowing multiple internal segments, port switching allows individual users to be connected to a specific LAN, providing very high levels of flexibility for network administrators. Like module switching, port switching allows all users on the defined segments to be interconnected using bridges or routers. Understanding Switching and Virtual Networking The ability to quickly connect any port to a different segment is ideal for networks in rapidly growing environments with frequent adds, moves, and changes. With port switching, reconfigurations can be done simply with network management software - no physical intervention at the station or hub is required. Port switching also allows a network administrator to respond to change requests at remote sites where there may be little or no network support staff, saving time and money. In some networks, certain computing resources or personnel are more prone to move within the organization. Port switching can be used selectively for these devices or people, while fixed network or module switching hubs are emploved for less transient equipment or people. Combining port switching with shared LANs and module switching can create a cost-effective solution for meeting specific network needs. LAN Switching Module and port switching provide static switching and allow network administrators to move entire modules or individual users among the internal segments within a multi- segment wiring hub. These switching technologies allow network administrators to easily reassign users to a given network segment to improve bandwidth utilization. A more powerful technology, LAN switching, offers an even greater degree of bandwidth enhancement and connection flexibility. LAN switching is achieved using a device that generally operates as a high-speed, multipart local bridge. Each port can be viewed as a separate LAN segment or ring. The most general purpose LAN switches allow each port to support either a single user or a large number of device addresses. This feature provides the network administrator with the ability to support either a single power user or an entire shared LAN segment on one port of the switch. LAN switches provide dynamic connections between port pairs in a fashion that resembles a telephone central office (CO) switch. As is also the case in a CO switch, LAN switches can supports multiple simultaneous connections. It is this characteristic that is mainly responsible for a LAN switch's ability to enhance effective bandwidth. Understanding Switching and Virtual Networking xref image = fig8.tif Figure 8: Ethernet LAN Switching In order to establish connections dynamically, a LAN switch must maintain a directory of addresses present on each port. This directory is often referred to as the forwarding table. The LAN switch learns MAC addresses by observing the source address of incoming packets on each port and builds forwarding tables based on these addresses. The switch intelligently forwards packets based on this table, 'ust as a bridge does. Beyond this, LAN switches can be quite different from standard bridges. Generally, LAN switches are capable of high forwarding rates, typically media speed, between sets of port pairs. Many LAN switches can support a large station population on each port, although some power users or servers may be the only device on a port. Supporting a single device on a port effectively allocates the full bandwidth of the segment or ring to the connected device. Using a LAN switch in this way represents the highest degree of segmentation possible. In recognition of this small but growing use of LAN switching, some LAN switch implementations now support only a single address (or a very small number of MAC addresses) on each port. This implementation allows some simplification to the design of the switch, reducing the cost per port but eliminating the general purpose solution. While LAN switches are technically feasible for all LANs such as Ethernet, Token Ring, and FDDI, they have been implemented primarily for 10 Mbps Ethernet LANS. Many users have already determined, and analysts increasingly agree, that providing 10 Mbps to a high- performance client or server device and sharing IO Mbps over an appropriately sized workgroup are sufficient for meeting most current throughput needs. In the coming years, it is expected that a few very high-end client applications, as well as an increasing number of high-performance servers, will require dedicated bandwidth in excess of 10 Mbps. LAN switches can also address these higher-speed requirements by either allocating multiple parallel network interfaces to a single server or complementing the Understanding Switching and Virtual Nctworking standard-speed ports with one or more higher-speed interfaces to the server (see Figure 8). To use parallel network interfaces, the server must be equipped with a special driver that is designed to take advantage of this feature. Use of a higher-speed interface to the server requires that the server be equipped with the appropriate NIC card and driver, and also assumes that the server is capable of sustaining relatively high throughput across the interface. Some popular high-speed interfaces include FDDI, TP-DDI, I 0OBase-T, 1 0OBase- VG, and ATM. xref image = fig9.tif Figure 9: Ethernet Switching with a High-Speed Server Interface Any of these high-speed interfaces can also be used for creating high-speed connections to backbone networks. This allows LAN switches to be interconnected at relatively high speeds in distributed or collapsed backbone configurations. ATM Because some LAN switches with a high port density are capable of generating traffic volumes that exceed the capacity of typical high-speed interconnects, such as 100 Mbps FDDI, there is increasing interest in backbone interfaces operating at even higher speeds. ATM may offer such a solution since it aims to provide standardized interfaces at 155 Mbps and 622 Mbps. Many network administrators view ATM as the ultimate solution to the network throughput challenge, but at too high a price (currently about $3000 per port) to be used as the sole network distribution technology. Although ATM was initially conceived as a technology for the wide area network, it has recently received much attention for use In local area networks for premises backbones. At present, ATM is rarely used as a client delivery technology since, unlike LAN switching which preserves the investment in the existing network infrastructure, ATM is not compatible with existing network components. However, a combination of ATM and LAN Understanding Switching and Virtual Networking switching provides an attractive solution for a high performance, scalable network - LAN switching provides a cost-effective client distribution mechanism while maintaining a significant percentage of the investment in the network infrastructure and ATM provides a high-speed LAN or LAN/WAN backbone (see Figure 10). xref image = fig10.tif Figure 10: Connecting LAN Su,itches using ATAI When to use LAN s itching LAN switching is primarily used in three applications: as a collapsed backbone solution in a fairly large corporate network, as a department backbone for increasing the available bandwidth, and as a high-speed workgroup solution. Collapsed Backbone Switch LAN switches can provide many times the effective bandwidth of an existing single- site network, currently supporting shared media or static switching technologies, in which overall traffic levels and the resulting congestion have reached a troublesome level. A LAN switch can provide many more segments than are offered by currently available module switching or port switching hubs. While module and port switching hubs usually support between two and five LAN segments or rings, LAN switching devices could support up to 100 Ethernet segments. In this application, the LAN switch functions as a high-performance multipol-t bridge/router in a collapsed backbone mode, but with less functionality and at a lower cost than the current generation of high-end bridge/routers (see Figure I 1). Understanding Switching and Virtual Networking xref image = fig11.tif Figure I 1: LAN S itch as a Collapsed Backbone The primary difference between a LAN switch and a multipart bridge/router is that the switch typically offers no WAN interfaces and may be limited to bridge-like connectivity between ports. If the corporate network is very large, encompasses multiple geographically distributed facilities, melds multiple LAN technologies, or requires rigorous security and traffic policy restrictions, then an enterprise class bridge/router will typically be required, possibly complemented by LAN switches at the department or division level Department Switch In very large and complicated networks that need the full services of an enterprise class bridge/router at the backbone level, LAN switches can be used to address the department or division level requirements in a cost-effective manner (see Figure 12). In this situation, LAN switching can be used to enhance the bandwidth available within the department or division, but still integrate easily into the corporate backbone network. Important characteristics in this configuration may include the port density provided, the range of backbone interfaces supported, the number of MAC addresses supported on a given port, and the level of management information that the unit can provide. Understanding Switching and Virtual Networking xref image = fig12.tif Figure 12: Department-Level LAN Switch Workgroup Switch LAN switches can also be deployed to serve the high performance requirements of a specific local workgroup. For example, a CAD/CAM department might be using high- performance workstations that frequently access servers. If these workstations were connected directly to the main corporate network, the level of traffic that they generated could have very noticeable negative impacts on the rest of the user population. By connecting the workstations and servers through a LAN switch, the performance of the CAD/CAM department can be improved, and the traffic they present to the corporate network can be limited to routine traffic such as e-mall and access to corporate data bases. While significantly lower than the per-port cost of FDDI (approximately $2000 per port) and ATM ($3000 to $4000 per port), the cost of basic LAN switching equipment today is still higher ($300 to $750 per port) than module switching (approximately $150 per port) and port switching (approximately $250 per port). Because of this capital investment, the decision to use a LAN switching solution must be carefully considered. To make the most informed decision, potential users of LAN switching should fully understand their intended applications. This understanding allows various vendors' design criteria to be compared and contrasted against network requirements. Understanding Switching and Virtual Networking LAN SWITCHING CRITERIA Due to its compatibility with existing network equipment, LAN switching is often the solution of choice for network administrators who need to reduce network congestion in the immediate future. Not all LAN switch solutions are alike, however. There are many factors to consider when choosing a LAN switching solution and they include: o the role of the switch o s Iystem architecture o performance characteristics o port speeds and standards supported o data forwarding techniques o Level 2 versus Level 3 switching o port address characteristics o flow and congestion control Role of the Switch The first criterion to consider when choosing a LAN switch is the role that LAN switching will serve in the network. As described earlier, LAN switching applications can roughly be divided into three primary applications - corporate, department, and workgroup. A LAN switch that will exist at the top of the corporate network hierarchy will likely require a broader range of features and capabilities than a product intended to address a small, high- performance workgroup application. Features that are non-essential at the workgroup level may be very critical at the corporate or enterprise level. For this reason, the per-port cost of a LAN switch targeted at the corporate network level will be higher than a unit designed specifically for the workgroup. While the corporate level product might easily be used at the workgroup level, it may not be the most cost-effective solution. System Architecture Once the intended role of the LAN switch is determined, it is important to choose a LAN switch architecture that will easily integrate into the planned or existing network. Corporate-level Soltitions If the intention of a corporation is to design the corporate network around a LAN switching architecture, then a high-end LAN switch will be the appropriate solution. This unit will most likely be a switching hub or modular switch that provides the following set of features and functions: o high port densities (typically up to 100 ports) o large number of MAC addresses per port (approximately 1000) Understanding Switching and Virtual Networking o high-speed backbone interconnect, such as FDDI o high-speed server interfaces o fault-tolerant features such as redundant power systems and backbone connections o sophisticated network management o integrated routing features to provide security, firewalls, and possibly WAN interfaces o complementary functions such as shared media repeaters and lower-speed asynchronous interfaces When choosing a switching hub implementation, it is important to consider the other technologies integrated into the hub, in addition to LAN switching. Examples include local routing, WAN routing, and lower-speed shared media functions (see Figure 13). Integration of these technologies into a single chassis provides not only a cost-effective single-device solution, but also a solution with a unified management scheme which can save time and money during device configuration and troubleshooting. xref image = fig13.tif Figure 1 3: Integrated Hub Solution Department-level Solutions Department-level switching solutions may be satisfied with a less comprehensive product. The switch may consist of a fixed configuration standalone unit or a LAN switching card integrated into a traditional wiring hub. Depending on the specific needs of the department and the requirements for backbone interconnection, the LAN switch may include the following: o low to medium port densities (5 to 48 ports) o hundreds of MAC addresses per port Understanding Switching and Virtual Networking o high-speed backbone interconnect o high-speed server interface(s) o fault-tolerant features such as redundant power systems and backbone connections o strong network management o complementary functions such as shared media repeaters and lower-speed asynchronous interfaces Workgroup-level Solutions The requirements for a workgroup-level LAN switch vary dramatically depending on the planned use. A modest workgroup switching scheme might be satisfied by a single small standatone unit or a low port density hub switch card. Some high-bandwidth workgroups might require medium port density (particularly if a single station is connected to each port), multiple high-speed server ports, and a high-speed backbone connection. A typical low-end workgroup switch may have the following characteristics: o low port densities (5 to 12 ports) o tens of MAC addresses per port o basic network management A typical high-end workgroup switch may have the following characteristics: o medium port densities (12 to 48 ports) o I to approximately I 00 MAC addresses per port o high-speed backbone interconnect o high-speed server interface(s) o strong network management Performance Characteristics LAN switch performance is an area that is often misunderstood. Ideal LAN switch performance is a function of the number and speed of the ports provided, the relationship between the speed of the low-speed ports and any high-speed ports, and whether ports operate in half-duplex or full-duplex mode. The actual performance of a LAN switch is also influenced by the particular hardware implementation that a vendor uses and the types of flow control and congestion avoidance techniques that are employed. LAN switch performance can be measured using a number of performance metrics including filtering performance, steady state and burst forwarding performance, and steady state and burst aggregate throughput. Most LAN switch vendors emphasize the areas in which their products perform the best. Understanding Switching and Virtual Networking A properly designed LAN switch will provide full rate filtering on all ports. This guarantees that all incoming packets, regardless of the arrival rate and the packet size distribution, will be analyzed as they are received. Throughput, which may also be referred to as the forwarding rate, can be specified on a per-port basis or a per-unit basis. Analyzed individually, each port should be capable of full forwarding at the maximum speed of its media type. Aggregate throughput, or the aggregate forwarding rate, is more difficult to analyze, especially if the switch supports both high- and low-speed interfaces. Compounding the difficulty is that fact that some vendors measure the steady state forwarding rate and others measure the burst forwarding rate. For a switch with all ports of the same speed, the steady state forwarding rate is measured assuming that all ports are paired off in groups of two. To make the measurement, all port pairs are activated simultaneously with one port in each pair sending data and the other receiving data. The measurements for all of the port pairs are then added. For example, the aggregate throughput for a I 6-port Ethernet switch would be 80 Mbps (8 pairs of ports, each transmitting at 10 Mbps). For a switch with one or more high- speed interfaces, the measurement is more complicated and is influenced by the number of lower-speed ports and the speed of the high-speed interface(s). The burst forwarding rate is measured over a short duration during which all ports can operate at full media speed and buffer any excess data that cannot be forwarded due to blocking of an output port. The duration of the burst must be short because the data buffers available to each port wilt typically be relatively small to avoid unnecessary delay. Port Speeds and Standards The most widely installed LAN switches provide support for 10 Mbps Ethernet. This is changing, however, as more units support one or more high-speed interfaces in addition to the lower-speed 10 Mbps interfaces. High-speed interfaces can be used to support server and high-speed client connections and/or to provide an interface to a high-speed backbone. The most popular high-speed interfaces include FDDI and TP-DDI, 100 Mbps Ethernet, and 155 Mbps ATM. Determining which high-speed interface is appropriate for a given application will depend on the installed backbone, the high-speed interfaces that can be supported by the servers and clients, and the performance level desired between the switch and the high speed client, server, or backbone. Data Forwarding Techniques v Another criterion to consider when choosing a LAN switch is the data forwarding techniques employed. The two most commonly implemented data forwarding techniques are store-and- forward and cut-through bridging. Fervent debates have been held discussing the merits and drawbacks of these two alternatives. The debates have focused primarily on the issue of latency through the switch, but switch latency is only one component of the total network delay. There are two other components that contribute to network delays - packet latency Understanding Switching and Virtual Networking through every device on the network including end stations, and packet errors. Po-cket latency is defined as the period of time that elapses between receipt of the first byte of a packet and the subsequent retransmission of that same byte. Packet errors contribute to network delays by creating the need for higher level software to retransmit data which increases overall network traffic. While delay through the switch is an important issue, limiting total network delays is critical in a switching environment supporting applications such as multimedia that can be sensitive to delays that either exceed maximum length or significantly vary over time. Store-and-Forward vs. Cut-Through Bridging A number of vendors have supplied products that use cut-through bridging as the forwarding technique. Cut-through bridging allows the beginning of a packet to be transmitted before the end of the packet has been received. This technique is used in an attempt to reduce switch latency. While this technique does reduce the latency measured through the LAN switch, it can actually add to the total network delay by forwarding packets before they have been checked for validity, which may allow packets with errors to propagate. This propagation of frames with errors can create a significant amount of unnecessary traffic on the LAN segments. The most widely implemented switch forwarding technique is full store-and-forward, which is implemented much like a MAC layer bridge. Packets are fully buffered internally and checked for validity before the forwarding process is initiated. The store-and-forward technique has three advantages over cut-through bridging: o Data integrity is assured. Runt, giant, and pygmy packets and packets with CRC errors are detected and not allowed to propagate. o Port speed mismatches are supported. Cut-through bridging assumes that data is arriving and departing at the same speed. When there are port speed mismatches, as is the case when an Ethernet port is switched to an FDDI port, cut-through bridging cannot be used. Store-and-forward techniques are required in this situation because complete packets must be forwarded. o MAC layer conversion is supported. Conversion from one LAN technology to another, for example Ethernet to FDDI, requires the complete buffering of a frame before it can be translated. In this case, cut-through bridging is not applicable and store-and- forward must be employed. Fragment-free Forwarding A third forwarding approach that has been recently implemented by some vendors is fragment-free forwarding. Fragment-free forwarding waits for the first 64 bytes of an Ethernet frame to be received before transmitting the data to determine if the frame meets the minimum frame size requirement. This approach eliminates the transmission of short packets and fragments of frames on the network, but still has the other limitations associated with cut-through bridging. Understanding Switching and Virtual Networking Prioritization A priority mechanism can be combined with store-and-forward processing to ensure unimpeded processing of high priority packet streams. Traffic generated by or destined for high priority devices can be given preferential access to available throughput and buffering space, providing the lowest possible packet latency and the best guarantee of error-free transmission. Level 2 vs. Level 3 Switching v The majority of products in the LAN switch classification operate at Level 2 of the OSI model, providing either store-and-forward or cut-through bridging operation at the MAC layer. These are sometimes referred to as Level 2 switches.A minority of vendors who classify their products as switches have implemented full network layer, or Level 3, routing on a per- port basis. The routing support is usually limited to a few important protocols such as IP, IPX, and DECnet. These products are sometimes referred to as Level 3 switches. While products that provide store-and-forward bridging and Level 3 routing are really best described as multiprotocol bridge/routers, these products are nonetheless included in most reports on LAN switching. The distinct advantages of LAN switches should be that they are simple to use and administer, provide an alternative to the complexity of multiprotocol bridge/routers, and are less expensive per port than multipart bridge/routers. The addition of routing on a per-port basis, in contrast to providing routing on a single backbone port, tends to violate each of these three principles. Routing does play an important role in large networks. A premises-wide LAN switching solution will typically employ routing to interconnect bridged domains. This approach creates firewalls that help eliminate the possibility of broadcast storms. Routing also provides the ability to convert from the predominant LAN technologies (usually Ethernet) to other technologies such as FDDI, WAN links, and ATM. There are two ways to implement the appropriate level of routing required in a network. Routing can be provided with a separate enterprise class bridge/router device or it can be integrated into the LAN switch. If routing is integrated into the LAN switch architecture, the simplest and most cost-effective implementation is to reserve routing for a few specialized ports on the LAN switch. This helps maintain the simplicity and low cost implementation of Level 2 LAN switches and will satisfy the requirements for routing in certain high-end applications. Understanding Switching and Virtual Networking Port Address Characteristics The maximum number of stations to be connected to any one port will determine how many addresses per port are required. Some vendors' products fix a maximum number of addresses that are available on each port. For example, the number of addresses supported per port on an Ethernet LAN switch may range between I and 1024. Other vendors provide a maximum number of addresses available in the unit, typically in the 256 to 16,000 range. Since storing addresses in a forwarding table that can be accessed at very high speeds requires a lot of high-- speed memory, there is a fairly strong correlation between the maximum number of addresses provided per port or per unit and the price of the LAN switch. The lowest cost implementation for the manufacturer is to provide a single MAC address on each port. In this case, the cost for the customer of connecting a single user Is truly the cost of an entire port on the LAN switch; there is no possibility of dividing the cost of the individual LAN switch port over multiple attached devices. This eliminates the ability of the LAN switch to support existing LAN segments and to support a growing network. Flow and Congestion Control Another criterion to consider is the flow and congestion control mechanism used in the LAN switch. Ideally a LAN switch should incorporate a non-blocking design which ensures that there is always enough throughput for each connected port to send or receive traffic from independent sources and destinations at full LAN speeds (e.g., at 10 Mbps for an Ethernet switch). But even systems designed to be non-blocking may experience output blocking. Output blocking occurs when there is not enough bandwidth to support the activity of a specific device. For example, large numbers of devices sharing a single server can cause output blocking. In this case, each transmitting device has full LAN speed access to the server, creating the need to send more than 10 Mbps (for an Ethernet switch) to the server. Figure 14 contrasts a non-blocking traffic pattern with a pattern that could introduce output blocking. xref image = fig14.tif Figure 14: Non-blocking and Output Blocking Traffic Patterns Understanding Switching and Virtual Networking Occasional output blocking is not uncommon in a busy network. Flow control options are designed to help guard against packet loss and performance degradation when output blocking occurs. One flow control option is to provide each switch with the ability to activate a back pressure algorithm. When in use, this algorithm causes hardware to sense and limit the potential for packet overrun by consuming bandwidth on the segment generating the excess traffic. For example, an Ethernet switch will send additional packets to the offending segment, causing the connected devices to defer transmission and limiting their transmission rates while the traffic load is high. Most packet loss is avoided and retransmissions at the upper protocol layers are limited. Another way to control congestion is to support the designation of individual devices as high or low priority. Data streams to or from high priority devices can be given preferential access to internal buffer space and throughput. This allows network administrators to lessen the likelihood of packet loss in critical applications. Full-duplex Ethernet Full-duplex Ethernet has also been implemented as an approach to congestion control. Full- duplex Ethernet makes use of the fact that there are two pairs of wires in star-wired 1OBase-T environments - one for transmitting and one for receiving. By installing NIC cards, hub ports, and server and client software that support full-duplex Ethernet, the effective bandwidth can be doubled. The primary advantages of full-duplex Ethernet are that it can co-exist with normal half-duplex Ethernet and make use of existing 1OBase-T wiring. The major disadvantage of full-duplex Ethernet is that it requires investment in new NIC cards, hub ports, and software. After evaluating the applications on a network, choosing the technologies that best address the reduction of network congestion, and building the network infrastructure, the next issue facing the network administrator is how to manage this complex network. The connection flexibility provided by LAN switching in particular creates the need for a way to represent devices on a network as functional, not physical, devices. This need has spurred the emergence of virtual networking. Understanding Switching and Virtual Networking VIRTUAL NETWORKING There is a lot of confusion about what virtual networking actually is or, more accurately, what it will be. Virtual networking, in simple terms, is a network management tool that will provide the functions necessary for managing a dynamic and complex network. Ideally, virtual networking will allow a user on a network to move to another building, work from home, or call in from a hotel without changing the configuration of the end station or creating the need for the network administrator to reconfigure any routers on the network. For a virtual networking tool to be effective, it will need an extensive graphical user interface (GUI) combined with well-developed SNMP agents in the managed devices. Vendors with experience in both network management tools and software-based networking devices will have an advantage over vendors lacking this experience. Specifically, virtual networking will allow network administrators to: o Create network groups or virtual LANs based on varying criteria o Allocate and protect network bandwidth o Control access to the network o Reduce the cost and complexity of management activities Creating Virtual LANs A necessary component in implementing virtual networking technology will be the ability to create virtual LANS. A virtual LAN is a collection of devices that have been grouped together logically based on their needs, traffic patterns, location, or any other criteria the network administrator defines. Devices in a virtual LAN may be on different physical LAN segments but will function as if they are directly attached to the same LAN. Network administrators will be able to create virtual LANs from a network management console rather than at a wiring closet, saving time and money. The ability to create virtual LANs will allow the network administrator to group users on an as-needed basis so that they can share information in an unrestricted fashion. Devices not explicitly placed in a particular virtual LAN will be placed in a default LAN which governs their communication capabilities. Standard chassis-based or external bridges and routers can be used to provide communication paths between virtual LANs (see Figure 15). Understanding Switching and Virtual Networking xref image = fig15.tif Figure 15: Virtual Perspective of a Network Allocating and Protecting Network Bandwidth Any switching technology that has to handle broadcasts or multicasts, including LAN switching and ATM-based LAN emulation, needs to deal with the issue of scalability. For example, unless restrictions can be placed on traffic flow, when a device sends a multicast it would be delivered to all other devices. As the number of devices increases, this could introduce multicast storms which could undermine the throughput enhancements that switching provides. Virtual networking will allows network administrators to limit the propagation of broadcast and -multicast packets. Broadcasts and multicasts will be exchanged freely within a workgroup or virtual LAN; they will not be sent outside a workgroup. Using routers instead of virtual networking to accomplish this would be more expensive and provide lower effective throughput. Controlling Network Access Network administrators will also be able to use virtual networking with bridges and routers to set up firewalls that define communication boundaries. Using a routing function, for example, a network administrator can provide every user on the network with access to electronic mail, but can restrict access to the server in the finance department to members of the finance department. Understanding Switching and Virtual Networking Reducing Cost and Complexity of Management In general, existing routers in a network will not allow a device to move from one physical LAN to another without either a change of the device's network address or a reconfiguration of some or all routers in the network. In large networks, considerable network administration time can be spent tracking these changes. Virtual networking will eliminate these costs by guaranteeing that all communications from the stations in a workgroup to a router always occur over the same interface to the router. Using virtual networking, a user could move to an office in another building without any need for the network administrator to reconfigure the network. This will help control the management overhead associated with moving, adding, or changing the deployment of stations in the network. It will also provide centralized access to devices, limiting the need for physical access to end-stations. Internal and external bridges and routers can also be used to connect multiple chassis and extend the scope of a network. This capability will be extremely powerful when combined with the ability of virtual LANs to span chassis. Coexistence of these additional internetworking devices with existing backbone architectures is also critical. CONCLUSION Most corporations have a variety of computing needs that require an equally diverge mix of technical solutions. For example, a single corporation may use module switching, port switching, LAN switching, FDDI, and ATM (see Figure 16). o The telemarketing group may only require word processing, electronic mail, and database access. Segmenting this group using internetworking and module switching provides them with access to the rest of the network, but separates them from backbone traffic. o A remote site may support a sales and support staff, but have no network administrator. Supporting this site with port switching allows the network administrator at the corporate headquarters to make changes to network connections through software. o The users in marketing communications running high-volume multimedia applications may best be supported using LAN switching. o The engineering department in a separate building may need to share 3-D images which may require a dedicated FDDI connection to the server in combination with LAN switching throughout the department. o Connecting the two buildings using ATM will provide a technology with enough bandwidth to handle the exchange of files between the marketing and engineering departments. Understanding Switching and Virtual Networking xref image = fig16.tif Figure 16: Example Corporate Network Choosing from the many technologies available today, including the various switching solutions, andplanning for future ones, such as ATM and Virtual Networking, present new opportunities and challenges for network administrators. When looking at a solution to increase throughput, several factors should be taken into account. These include not only the applications being supported but also price, performance, flexibility, and compatibility with existing equipment. In addition to these qualities, it Is also important to consider the following factors. Breadth of Product Offering Since multiple technologies can be employed to solve throughput problems, the ideal vendor will support several of them, allowing for the flexible re-deployment of products. A single- vendor solution also tends to reduce the cost of network management training, provided the vendor supports its spectrum of products with a unified network management solution. Preservation of Investment Because network administration budgets are generally constrained, it is important to examine today's products for migration paths to new technologies. A vendor should have a proven track record of providing solutions that are easy to upgrade, not forklift solutions. Understanding Switching and Virtual Networking Comprehensive Network Management As networks grow and support staff and budgets shrink, the ability to respond quickly and easily to change requests and network faults becomes even more important. Extensive network management applications and rich SNMP agent implementations can provide the tools necessary to meet these challenges. As demonstrated, choosing the best combination of technologies for a particular network can be a challenge. Choosing a vendor that supports all available networking technologies, particularly within a single chassis, greatly reduces that challenge and minimizes the risks. Xyplex is unique in its ability to offer this combination. THE XYPLEX SOLUTION The Xyplex product line, including the award winning Network 9000 Routing Hub, offers state-of-the-art solutions to the throughput challenge. Specifically, the Xyplex product line offers the following advantages. Software-based Architecture The software-based architecture of Xyplex products ensures seamless technology integration, superior functionality, common human interfaces, and detailed device management. It also allows consistent functionality, syntax, and 'look and feel' in both the Network 9000 and standalone devices, greatly reducing the operation and training costs as new component are added to the network. Core Technology Ownership Xyplex designs, develops, and manufactures its own media access, switching, routing, and interface equipment. This fact ensures not only seamless technology integration but also a more thorough and responsive support organization. Expertise in these networking technologies provides a strong foundation for migrating to future technologies as wiring- based and software-based technologies continue to merge. Compatibility Because Xyplex has extensive expertise in routing technology, it is in a unique position to interface with other manufacturer's backbone routers. Xyplex understands routing protocols and consistently provides products that tightly integrate with software-intensive products from other companies. Understanding Switching and Virtual Networking Breadth of Product Offering Xyplex owns a broad spectrum of core networking technology. We provide LAN connectivity for local users, dial-in connectivity for remote users, and wide area connectivity for branch offices. Preservation of Investment Xyplex, alone among intelligent hub vendors, has a single design, the Network 9000, that spans the spectrum from cost-effective repeater-based media connectivity to software- intensive LAN switching and integrated routing with ATM support. Comprehensive Network Management The award-winning ControlPoint network management application provides extensive SNMP-based management for all Network 9000 components. A point-and-click graphical user interface makes it easy to move devices in module switching, port switching, and LAN switching environments. ControlPoint will support Xyplex's Virtual Networking Environment (VNE) which among other features wilt provide the capability of virtual LANs to span chassis. The Network 9000 Routing Hub is designed to be easily upgraded to support current and future technologies, including 100 Mbps Ethernet and ATM. Network administrators can install a Network 9000 today and be assured that their investment in Xyplex equipment will carry them into the future of networking technology. Understanding Switching and Virtual Networking GLOSSARY 1OBase-T 1OBase-T is an approved proposal by the IEEE 802.3 committee for an industry standard enabling 10 Mbps Ethernet LAN traffic to be transported over 24-gauge unshielded twisted pair wiring. 100Base-T 100Base-T is an IEEE 802.3 proposed standard for enabling 100 Mbps Ethernet LAN traffic to be transported over 24-gauge unshielded twisted pair wiring using the CSMA/CD access method. 100Base-VG 100Base-VG is an IEEE 802.12 proposed standard to specify the transmission of Ethernet LAN frames at 100 Mbps over unshielded twisted pair and fiber media. ATM (Asynchronous Transfer Mode) ATM uses 53-byte fixed-cell relay transport technology to provide high-speed (155 Mbps and higher) local and enterprise-wide data transport. Back pressure algorithm The back pressure algorithm is a flow control mechanism used in LAN switching that causes hardware to sense and limit the potential for packet overrun by consuming bandwidth on the segment generating the excess traffic. Bridge A bridge is a device that connects local or wide area networks at the data link layer. Broadcast Broadcast packets are sent to all stations in a network. Category 3 cabling The Electronics Industry Association/Telecommunications Industry Association (EIA/TIA) 586 standards specifies commercial building telecommunications wiring. Category 3 wiring is unshielded twisted pair (UTP) cable specified by the EIA/TIA 586 standard for speeds up to 10 Mbps and it is the minimum cable required for 1OBase-T Ethernet networks. Category 5 cabling Category 5 wiring is unshielded twisted pair (UTP) cable specified by the EIA/TIA 586 standard for speeds up to 100 Mbps, but 155 Mbps is also possible. CSMA/CD (Carries Sense, Multiple Access with Collision Detection) CSMA/CD is the IEEE 802.3 access method. Each network device waits until it senses that the network is not busy before transmitting and detects possible collisions with other data that may occur after transmission begins. Understanding Switching and Virtual Networking Cut-through bridging Cut-through bridging is a data forwarding technique used in LAN switches that allows the beginning of a packet to be transmitted before the end of the packet has been received. Data link See Level 2. Departmental LAN A departmental LAN is a network used by a small group of people working toward a similar goal. Its primary function is to share local resources, such as applications, data, and printers. Dynamic switching Dynamic switching establishes connections on an as-needed basis. LAN switching and ATM switching fall into this category. Ethernet An Ethernet network is a 10 Mbps baseband LAN that uses the Carrier Sense, Multiple Access with Collision Detection (CSMA/CD) media access method. 802.3 is the IEEE specification for an Ethernet network. FDDI (Fiber Distributed Data Interface) FDDI is the American National Standards Institute (ANSI) X3T9.5 specification for a 100 Mbps interface for fiber optic networks configured in dual counter-rotating rings. Firewall A firewall is an impermeable barrier through which broadcasts and other types of packets cannot pass. Routers, not bridges, are used to create firewalls. Fragment-free bridging Fragment-free bridging is a forwarding technique used by Ethernet LAN switches that waits for the first 64 bytes of an Ethernet frame to be received before transmitting the data to determine if the frame meets the minimum frame size requirement. IEEE 802.3 The Institute of Electrical and Electronics Engineers 802.3 specifies the physical and medium access control (MAC) standards for Ethernet networks. IEEE 802.12 IEEE 802.12 Is a proposed standard for 100 Mbps transmission using a shared media MAC standard called demand priority. Demand priority prioritizes traffic into high and low priority. Internetwork An internetwork is a collection of several networks that are connected by bridges and routers so that all users and devices can communicate with each other regardless of the network segment to which that are attached. Understanding Switching and Virtual Networking LAN switching LAN switching is a packet transfer mechanism used to provide full bandwidth interconnection for all LAN segments or devices connected to the switch. Latency Packet latency is the period of time that elapses between receipt of the first byte of a packet and the subsequent retransmission of that same byte. Level 2 Level 2 refers to the second layer, or data link layer, of the OSI model. It defines how data is packetized and transmitted to and from each network device. It is divided into two sublayers - medium access control (MAC) and logical link control (LLC). Level 3 Level 3 refers to the third layer, or network layer, of the OSI model. It governs data routing. MAC layer The MAC layer is the lower sublayer of the data link layer of the OSI model and governs access to the transmission media. MAC layer address The MAC layer address is a unique identification code used by the MAC layer to identify devices on a network. Module switching Module switching describes the ability to connect all the devices attached to a given module in a hub to one of the multiple internal physical LANS. Multicast Multicast packets are single packets that are copied to a specific subset of network addresses. In contrast, broadcast packets are sent to all stations in a network. Multimedia Multimedia is the incorporation of graphics, text, and sound into a single application. Network layer See level 3. NIC (Network Interface Card) A NIC is an adapter board that provides the physical connection between a computer and the network medium. OSI model The Open Systems Interconnection (OSI) model is the seven-layer, modular protocol stack defined by the International Standards Organization (ISO) for data communications between computers. The seven layers are: physical, data link, network, transport, session, presentation, and application. Understanding Switching and Virtual Networking Port switching Port switching describes the ability to switch users on a per-port basis among the internal LANs within a single chassis. Protocol A protocol is a standardized set of rules for establishing and controlling data transmissions, including formatting, timing, sequencing, and error checking. Router A router Is a device that connects LANs at the network layer and supports protocols required for the filtering of packets. Server A server is a computer that provides shared resources to network users. A server typically has greater CPU power, memory, cache, disk storage, and power supplies than a computer used as a single-user workstation. Static switching Static switching describes connections that must be made and changed under network management control. Module switching and port switching fall into this category. Store-and-forward bridging Store-and-forward bridging is implemented much like a MAC layer bridge in that packets are fully buffered internally and checked for validity before the forwarding process is initiated. Token Ring Token Ring is the IEEE 802.5 specification for a 4 Mbps or 16 Mbps network that implements a logical ring with a physical star topology. Token Ring uses a token-passing access method in which the transmitting station must possess an electronic token before it may send its packets to the network. TP-DDI (Twisted Pair-Distributed Data Interface) TP-DDI is a protocol, ratified by the American National Standards Institute, that converts 100 Mbps FDDI protocols to copper wiring. UTP (Unshielded Twisted Pair) UTP cabling consists of a pair of foil-encased copper wires twisted around each other. Virtual LAN A virtual LAN is a group of devices that appear to be connected by a multipart bridge, sharing data and multicast information in a closed environment. Virtual networking Virtual networking is a network management tool that allows the separation of the physical aspects of a network's topology from the services delivered. For further information please call ph: (07) 393 1933 fax: (07) 391 4143 email: netarch@kraken.itc.gu.edu.au Regards G. Croker NetArch - Brisbane