AT&T makes Ethernet switching as easy as a "Seabreeze"FOR RELEASE TUESDAY, AUGUST 22, 1995
BERKELEY HEIGHTS, N.J. -- AT&T Microelectronics has made it possible to design a cost-effective, expandable 12-port, four-segment Ethernet switching system which enables users to reconfigure port-switched hubs virtually on the fly.
With the combination of the newly-introduced ATT1S04 Seabreeze switching device and three ATT1RX04 Forespar quad single-port repeaters, system designers can now offer customers flexible network management features like per-port switching and port mobility.
"The Seabreeze switch is an ideal building block because its scaleable design allows it to be cascaded to support up to 192 ports among multiple segments," noted Clarence Joh, AT&T's Ethernet IC marketing manager. "It solves the OEM's concerns about time-to-market while matching the user's investment closely to current, as opposed to future needs."
While governing collision detection and data-path switching for each segment on a port-by-port basis, the Seabreeze device enables network managers to use software to reassign a port to a different segment, thereby eliminating the need for hardware changes.
Per-port segmentation and port mobility improves network management by enabling the network administrator to better balance traffic loads, set aside certain segments for redundancy and reliability -- all without costly hardware moves and rearrangements. Additional management features are possible through the Forespar's microprocessor interface, providing access to status and configuration registers, on-chip security options, and a packet postprocessor with an associated 32-bit event counter for statistical monitoring. Moreover, AT&T's hardware-based patented security feature is a selectable option for automatic protection from eavesdropping and intrusion.
The new Seabreeze switching device helps further reduce design costs by integrating miscellaneous board-level circuitry. Among on-chip functions are address decoding for three Forespar repeater devices, powerup reset and a software-controlled hardware reset to the repeaters. The Seabreeze device conditions a system-level hardware reset to the Forespar repeater and also indicates completion of any reset operation.
Packaged in a 132-pin bumpered plastic quad flat pack (BQFP), the ATT1S04 Seabreeze chip is available now at $9.15 each in 10K quantities.
For product literature, customers may call the AT&T Microelectronics Customer Response Center, 1-800-372-2447, Dept. P65 (in Canada, 1-800-553-2448, Dept. P65); fax number +1-610-712-4106 (especially for callers outside of North America); or write to AT&T Microelectronics, Room 21Q-133BA, 555 Union Boulevard, Allentown, Pa., 18103.
AT&T Microelectronics offers a full line of high performance integrated circuits, electronic systems and optoelectronic components for applications in network computing, telecommunications, wireless and messaging products and multimedia workstations.
Multi-port ICs make Ethernet switches cost effective for mainstream use
Local Area Network (LAN) switches are slowly moving from specialty to mainstream status due to their ability to extend the functionality and baseline capacity (number of nodes) of a LAN. Unfortunately, the low-cost, widely-available ICs used to implement the single-port LAN interface in workstations or passive hubs aren't cost effective for use in multi-port active LAN switches. Therefore switch manufacturers require an entire new class of VLSI devices to continue and drive down the cost of switching technology.
Ethernet has come a long way from the early days of bus topologies using shared coaxial media. First, the computing community devised the 10BaseT Ethernet derivative that took advantage of both the more easily maintained star topology and lower-cost, twisted-pair cabling.
Both, coax and 10BaseT Ethernet installations, however, share a similar problem -- bandwidth limitations. Both topologies rely on shared media and use collision detection and rebroadcast techniques as a media-access scheme. As the number of users on a network grows more collisions occur causing more data packets to be broadcast multiple times thereby resulting in degraded performance.
One way to increase performance in a LAN is to move to faster communication schemes. The ongoing 100BaseT and 100VG efforts, for example, are both techniques that extend Ethernet transmission speed from 10 Mbits/sec to 100 Mbits/sec. AT&T, in fact, supports both new standards.
But faster data transfers don't get at the crux of the shared-media limitations. Faster speeds certainly enable applications requiring rich media streams such as compressed digital video. But faster speeds alone don't eliminate congestion on the shared media and this congestion degrades performance regardless of maximum data transfer speed.
Ethernet switches are the best solution for maximizing LAN bandwidth. The heritage of the LAN switch is the telephone switch. A series of switches maintain a direct physical connection between two parties for the entire duration of a telephone call. The connected parties have the full use of the bandwidth supported by the connection during the call.
A switch that could establish a dedicated link between any two nodes on a network would be ideal but impractical. Moreover, data flow on an Ethernet LAN is broken into packets or frames so a pure circuit switch isn't required to effectively boost capacity. Instead, so-called frame and segment switches are two different ways to solve the bandwidth problem using switching technology.
Frame switching was the first Ethernet switching technology and was introduced by Kalpana and others in 1990. In a frame switching environment, each workstation or node connects directly to one port of the switch. The workstation Ethernet interface operates just as if it were connected to a shared-media 10BaseT topology, except, the dedicated switch port eliminates collisions. The workstation node can broadcast a data frame anytime it is ready.
In the frame switch, the switch port is responsible for receiving the data frame and sending the frame to the appropriate destination node. Each port of the frame switch, therefore, must integrate MAC (media access control) layer functionality because the port must be capable of examining the incoming frame and extracting the destination address.
Frame switches use several techniques to move a packet along to the destination. The most common method is called store-and-forward. In a store-and-forward switch, the port receives the entire incoming frame, checks the frame for errors, extracts the destination address, and forwards the packet to the destination node.
To eliminate the latency of waiting for the entire packet to arrive before forwarding the frame, some vendors implement a cut-through technology. Essentially, the frame switch extracts the destination address from the header of the frame and begins the forward operation immediately. While this technique reduces latency, it can result in the transmission of an error-laden frame.
Regardless of the frame-forwarding scheme, all frame switches share the same basic set of features and characteristics. The key feature of the switches is the ability to add LAN capacity by adding ports without degrading performance as long as the bandwidth of the switch isn't exceeded. The switches typically use RISC processors, a fast backplane, and a large shared-memory array to process packets.
The most significant drawback of frame switches is cost. The RISC processors and memory arrays, and the need for a MAC in every port make the switches relatively expensive -- too expensive for many mainstream LANs.
Segment switching takes a different tact to solve the bandwidth problem. Rather than eliminating the shared-media scheme of connecting nodes, segment switches attempt to reduce collisions to a negligible level by breaking the overall network into small segments. Within each segment, the nodes still rely on shared media and collision detection. But the system administrator ideally defines the segments so that most of the traffic on each segment is local to the segment and collisions aren't a problem.
Segmentation is a widely tested concept. System administrators have been segmenting Ethernet LANs for years, albeit by manually connecting patch cords in a wiring closet. The administrator would physically connect all workstations within a segment to the same passive hub, and messages between hubs would be handled by bridges.
Historically, the downfall of segmenting LANs was the manual effort involved. Administrators typically provided segments based on departmental boundaries such as accounting, marketing, or engineering. Problems would arise when a segment should more logically be formed in a different manner -- around a project team for example.
Consider a team working on a new product that might include some engineers, a marketing person, someone from accounting, a purchasing agent, and a manufacturing engineer. Without segment switching, different physical locales and the cost of manually wiring segments made it impractical to physically connect the project team members from different departments.
A project team with regular communication needs and with members connected to different LAN segments, could increase traffic and degrade performance throughout the corporate LAN. For example, if the team member from accounting broadcasted a new budget to everyone on the team, the broadcast would create traffic on all of the segments rather than a single segment.
Remotely controllable switches made remote management and configuration of a segmented LAN possible. The earliest segment switches, however, still had limitations. For example, the cost of the repeaters that broadcast an Ethernet packet across the shared media made it impractical to handle the segment switching on a per-port basis. Instead, a group of four or more nodes would be multiplexed through a single repeater making segmenting possible on a small group basis.
Per-port segmenting and virtual LANs
Newer VLSI ICs, however, now make per-port segmenting or micro-segmenting cost feasible. To drive the price per port down, for example, AT&T has introduced ICs that support multiple channels including a repeater dedicated to each channel. These developments have led segment-switching proponents to coin a new term -- virtual LANs -- describing the ability to connect any user to a dedicated LAN segment on an ad hoc basis. The users connected to each segment have access to reserved bandwidth by being physically switched onto the segment. Should the organization change, the administrator can remotely reconfigure the segments.
Specifically, two VLSI ICs from AT&T provide switch vendors with the capabilities required to build segment switches that can handle per-port switching. First, the ATT1RX04 Forespar announced as the T7204 in 1994 integrates four independent Ethernet channels. Each channel includes a transceiver, a repeater, and a NRZ (non return to zero) backplane interface.
Second, the ATT1S04 Seabreeze includes 12 NRZ ports and can independently switch each of the 12 ports to one of four circuit segments. The Seabreeze is being announced in August, 1995.
The Forespar IC implements a comprehensive set of features that allow switch vendors go beyond offering simple port switching capabilities. The front-end transceiver interface, for example, directly supports 10BaseT twisted pair and external AUI (attachment unit interface) transceivers. The AUI compatibility will allow both coaxial or fiber media to connect to a segmented network. Each transceiver also supports smart squelch, link integrity detection, long line length, and automatic polarity reversal direction and correction. The back-end NRZ backplane interface supports data, collision, carrier sense, and clock signals.
The repeater core is the heart of the Forespar chip and implements a rich feature set including:
The Forespar IC also includes an AT&T-patented hardware security controller, port configuration and status registers, management statistic registers and counters including support for RMON statistics, and status LED controllers. These functions are linked into each repeater channel and provide the basis through which a switch vendor can offer robust network management capabilities. A microprocessor interface allows for remote monitoring and configuration of each port.
The hardware-based network security controller handles automatic eavesdropping prevention, and intrusion detection and control. The controller supports one source address per port, and can learn the source address automatically or under programmable control. The security features work with broadcast and multicast packets. And the intrusion detector can lock the last source address for management notification.
In addition, the ATT1RX04 optionally supports two universal media ports for 10BaseT, AUI, 10BaseFL, and 10BaseFB. For 10BaseFL and FB, the twisted-pair and AUI interface logic is bypassed to reduce the latency through the repeater core.
A single ATT1S04 Seabreeze IC is the ideal companion to a set of three Forespar ICs. Together, the four chip set can implement a complete 12-port, 4-segment switch.
The Seabreeze is an equally robust complement to the Forespar, providing data path switching, collision detection, and address decoding for a glueless interface to the quad-repeater ICs. The Seabreeze also supports system-level reset functions including power-up reset, and software-controlled hardware reset of attached Forespar repeaters.
Switch vendors can also cascade Seabreeze ICs to design hubs supporting more than 12 ports. A 4-bit ID allows for expansion to 192 ports. Moreover, the IC also includes provisions to support more than four segments when cascaded.
The ATT1RX04 Forespar and ATT1S04 Seabreeze provide switch vendors with the first truly cost-effective solution for reconfigurable, port-switched hubs. The ICs support flexible configurations, and allow the vendors to meet the intense time-to-market pressures that are prevalent in today's computer industry.
For more information, reporters should contact:
KEYWORDS: port_switched, lan, 10baset, 100vg, att1s04, forespar, repeater