Switched networks extend life of today's internetworks.

Together, frame switching and asynchronous transfer mode (ATM) cell switching offer the scaleable performance and flexible connectivity required to take internetworking into the next century.

Editor's note: This article is the first in a series of three that will provide a brief tutorial on new networking building blocks and architectures, as well as strategies for evolving current infrastructures into the new switched virtual networking paradigm.

Seemingly overnight, local area networks (LANs) have evolved into enterprise internetworks, the communications backbones of most modern organizations. But now internetworks are becoming victims of their own success. Continued growth and increasing usage are pushing the limits of traditional technology. Changing business practices and new applications require levels of performance, flexibility and reliability that today's internetworks cannot provide. Solutions are needed that deliver more bandwidth, carry multimedia traffic and simplify network administration.

Switch-based technologies promise to extend the useful life of today's internetworks and enable new architectures that will meet tomorrow's needs. Together, frame switching and asynchronous transfer mode (ATM) cell switching offer the scaleable performance and flexible connectivity required to take internetworking into the next century.

The limitations of today's internetworks have motivated the development of several new technologies. Some are simply faster versions of traditional building blocks, but two--frame switching and cell switching--bring dramatic new capabilities to internetworking.

Frame switching multiplies LAN performance at low cost while protecting user investment in end-station software, network interface cards (NICs) and building wiring. ATM cell switching provides additional scalability and also delivers the quality of service (QoS) required by multimedia applications. Virtual networking, an integral feature of most switching products, uncouples network addressing from physical topology, allowing a more efficient alignment of internetwork traffic flows with user work patterns. Specifications from the Internet community and the ATM Forum standardize the use of ATM in internetworks, ensuring frame-to-cell interoperability and multivendor compatibility.

High-Speed LANs

One approach to improving network performance is simply to install faster LANs. Many building and campus networks are already using 100-megabit per second (Mbps) FDDI. However FDDI is too expensive for general use as a desktop link, so two new 100-Mbps LANs, fast ethernet and 100VG-AnyLAN, have been introduced.

To date, 100VG-AnyLAN has only limited vendor support. It uses a new bandwidth allocation scheme that makes it incompatible with traditional LANs. In contrast, fast ethernet is compatible with conventional 10-Mbps ethernet and has gained wide support from networking vendors.

Fast ethernet (100BASE-T, is an extension of standard 10BASE-T ethernet) supports multiple media types and the same cabling installed for 10BASE-T ethernet can often be used for fast ethernet. Data can move between 10BASE-T and 100BASE-T LAN segments without protocol translation. A server connected at 100 Mbps integrates seamlessly with 10-Mbps clients.

Fast ethernet costs less per port than FDDI and is beginning to see use as a "fat pipe" to servers and power users and in downlinks to routers and switches. Compatibility with traditional ethernet allows the performance benefits of 100BASE-T technology to be integrated into an existing 10BASE-T network on an as needed basis, alleviating specific bottlenecks and evolving gradually to widespread deployment. NICs that work at both 10 Mbps and 100 Mbps afford a cost-effective upgrade path for newly installed user workstations or PCs.

Frame Switching

Frame switches, also known as LAN switches, offer the performance boost of LAN segmentation without consuming router ports and without requiring new NICs. Moreover, as ports are added to a frame switch, its overall throughput increases and network performance scales gracefully. Frame switches are similar to bridges. As each LAN frame enters a switch, its destination MAC address is compared with a table of previously learned addresses and the frame is sent out the proper port toward its destination.

But unlike a conventional bridge, which transfers packets by way of shared memory or an internal bus, a frame switch is built around a high-speed switching fabric. Application-specific integrated circuits (ASICs) in the switching fabric and at other key points in the device deliver higher throughput, lower latency and lower cost per port than traditional products.

Replacing a shared-media hub with a frame switch divides the LAN into multiple segments and gives each segment full LAN bandwidth. As with router-based segmentation, overall throughput is multiplied by the number of switch ports. Traditional LAN interfaces are used at each switch port, protecting user investment in wiring and NICs and ensures that existing, applications and network operating systems run unchanged.

Few organizations require full LAN bandwidth to every desktop, so the most common place to install a frame switch is behind a shared-media hub. The hub concentrates building wiring to form multi-user LAN segments, and the switch interconnects the segments at high speed. Modern chassis-based hubs include both shared-media and frame switch modules in the same enclosure.

Cell Switching

Frame switching augments LAN performance at the edges of a network and also can be used to relieve backbone congestion. But frame switching is ultimately limited by the speed and architecture of attached LANs. For links over 100 Mbps or for multimedia quality of service, ATM cell switching is required. ATM scales to gigabit-per second speeds, carries voice and video in addition to data and guarantees quality of service for all forms of traffic.

Cell switches differ from frame switches in two important ways:

1. Cell switches move cells rather than frames. Frames can vary in length from dozens to thousands of bytes, but ATM cells are always 53 bytes long.

2. Network frame traffic is connectionless, while ATM cells travel over virtual circuits. (Note: Not all frame traffic is connectionless. Frame relay, for example, moves frames over virtual circuits in a manner similar to ATM.)

In a traditional internetwork, each frame is a separate datagram, forwarded from point to point without a preordained path or guaranteed network resources. In an ATM network, each session is like a telephone call. Cells do not flow until an end-to-end path (i.e., a virtual circuit) has been laid out and network resources have been reserved.

These differences give ATM switches capabilities that cannot be achieved with frame switches. Short cell length keeps switch latency to a minimum. Very little time is required to pass a cell into and out of a switch. Fixed cell length and pre-allocated virtual circuits make cell processing simple and deterministic. Simplicity translates to highly scaleable performance. Determinism and low latency enable cell switches to meet stringent quality of service requirements for multimedia applications.

Many networks will include ATM LANs for workgroups with exceptional bandwidth or QoS requirements, such as radiology, but the dominant use of cell switches will be in the backbone of the enterprise network. As bottlenecks are cleared at the edges of the internetwork, the demand for backbone bandwidth will rise rapidly. ATM cell switches can meet the demand and scale as traffic continues to grow.

Frame-to-Cell Switching

The merger of frame and cell switching in high-performance networks has spawned another new building block, the frame-to-cell switch. Like a frame switch, this device interfaces directly to traditional LANs and is transparent to existing applications and network operating systems (NOSs). Like a cell switch, it sends traffic across the backbone as ATM cells. To do this, the frame-to-cell switch translates LAN frames into ATM cells and back again.

ATM Virtual Networking

Frame and cell switching facilitate LAN segmentation and enhance network performance, but they can also aggravate old problems. Without routing structure, a switched internetwork behaves like a bridged LAN. High-speed switching accelerates broadcast congestion and there are no security barriers between workgroups.

The traditional solution would be to make every switched segment a separate subnetwork, but that would make the internetwork impossible to manage. Every move, add or change would require address reconfiguration. Network administration would become a worse bottleneck than network performance.

Virtual networking resolves this dilemma by creating a logical structure independent of the physical internetwork. The performance benefits of LAN segmentation are retained while the logical overlay controls broadcast traffic and creates secure communities of interest. Traffic patterns reflect work flow rather than network topology. Network administration is simpler and more responsive to business re quirements.

Router-based bridge groups are another form of virtual networking. Multiple LAN segments are bridged within a backbone router to create virtual LANs (VLANs). Everyone in a VLAN shares a subnetwork address, which simplifies network administration and facilitates intra-workgroup communication. Routing between bridge groups maintains security and contains broadcast traffic.

Virtual networking is even more powerful when coupled with switching. Frame-to-cell switches are deployed on each floor, with an ATM switch and a router in the backbone. (Similar examples could be drawn using only frame switches, only ATM switches or a combination of various switches.) End stations are assigned to VLANs as dictated by workgroup organization, rather than wiring layout.

With virtual networking, the internetwork becomes "flatter." The number of subnetworks decreases because subnetwork addresses are assigned to virtual LANs rather than physical segments. Address administration is simplified because there are fewer subnetworks and because end stations can move between segments in the same VLAN without address reconfiguration.

Networking with ATM

The combination of frame and cell switching in an internetwork raises new architectural issues. How should connectionless network traffic be carried over ATM virtual circuits? How should islands of ATM fit into the overall network? How should LAN-attached end stations interoperate with ATM-attached servers?

There are currently three standard approaches to the use of ATM in internetworks as defined by the ATM Forum's Request for Comments (RFC). Two of the RFCs--multiprotocol encapsulation and classical IP over ATM--define standard methods for exploiting the scalability of ATM in today's internetworks, but both are limited in their application.

LAN Emulation 

The third RFC, LAN Emulation or LANE, establishes a standard for hybrid frame/cell switched LANs supporting multiple protocols. LANE provides transparent bridging of traditional legacy LANs (e.g. token ring or 10BaseT ethernet) over an ATM backbone. It also allows LANattached end stations (e.g., personal computers and workstations) to interoperate with ATM-attached devices such as database servers. With LAN emulation, network managers can apply ATM to network hot spots or implement an ATM backbone, while protecting their investment in existing LAN equipment and applications. LANE clients reside in ATM-attached bridges, routers, switches and end stations. They translate frames into cells (and vice versa) and steer connectionless LAN traffic onto the correct ATM virtual circuits. The ATM Forum has designed LAN emulation carefully so that LANE clients and ATM interfaces can be added to existing devices (e.g., LAN servers) without disrupting installed applications or network operating systems. Adherence to the LANE specification protects user investment in application hardware and software and ensures interoperability in multivendor internetworks.

Julie Ford is the healthcare program manager for Bay Networks, Inc. Billerica, Mass.

Julie's next Networking Strategies article will appear in the November issue. For questions or comments regarding the technology discussed in this article, contact Julie Ford at [phone].