Steve Bartolutti,
At&T Network Systems
During the past several years, there have been significant developments in computing and communications technologies. Although cabling installers primarily deal with the physical layer of such technologies, they should also know a little about the higher layers, so they can talk intelligently to the network managers who are their customers.
Consider, for example, the issues surrounding asynchronous transfer mode technology. ATM for local area networks is very much an infant technology with all of the associated start-up concerns. Although this technology for local area networks holds tremendous potential, there are several reasons why it will be a few years until ATM used in this capacity becomes a reality.
A number of short-term upgrade alternatives can bridge the gap to ATM. Each option has advantages and disadvantages, depending on application, and no single, optimum short-term solution exists for all circumstances.
Faces of Ethernet
Given that Ethernet is the most widely accepted and cost-effective form of computer networking today, it is not surprising that network development has been focused on creating a better Ethernet. From the various 100-megabit-per-second Fast Ethernet alternatives, to switched- and full-duplex Ethernet, and isochronous Ethernet, everyone is trying to retain Ethernet`s look and feel.
These Ethernet alternatives are generating much interest, because there are more than 40 million Ethernet connections worldwide--and the number is projected to grow. There is an impetus to develop a local area network implementation that looks and feels like the network to which users are currently connected.
Including Fast Ethernet, there are three 100-megabit-per-second alternatives under development. The first two, generically referred to as 100Base-T, are unshielded twisted-pair approaches being developed by Working Group 802.3 of the Institute of Electrical and Electronics Engineers. They are supported by the Fast Ethernet Alliance, an association of equipment vendors committed to this approach. The third is a multimode- fiber approach.
Of the two 100Base-T options, one--100Base-TX--operates over Category 5 unshielded twisted-pair cable, while the other--100Base-T4--operates over Category 3 UTP. Both preserve the carrier sense multiple access/collision detection medium access control layer that is unique to Ethernet.
100Base-TX can be considered 10Base-T scaled up by a factor of 10. This two-pair approach that supports 100-meter lobe lengths uses 4B/5B block-encoding and MLT-3 bit-encoding algorithms already being used for TP-PMD, or twisted-pair physical medium dependent. At a per-port price of just under twice that of 10Base-T, but with 10 times the network bandwidth, it is generating much interest among users.
The other UTP approach, 100Base-T4, supports 100-meter Category 3 UTP lobe lengths using all four pairs in the horizontal cabling--three for data transmission (at approximately 33 Mbits/sec each) and the fourth for handling collision detection. Utilizing non-return to zero bit encoding and a newly introduced 8B/6T block-encoding approach, 100Base-T4 was developed to provide a true Ethernet alternative to 100VG-AnyLAN.
The third IEEE 802.3 alternative is 100Base-FX, a multimode fiber implementation of Ethernet at 100 Mbits/sec. Using the same encoding approach as 100Base-TX, it runs over two fibers.
A pseudo-Ethernet approach, 100VG-AnyLAN, is also being developed by newly formed IEEE Working Group 802.12. Originally proposed to the IEEE as 100Base-VG by Hewlett-Packard Co. and AT&T Microelectronics Division, this alternative is also being supported by a group of approximately 25 companies known as the 100VG-AnyLAN Forum.
100VG-AnyLAN carries Ethernet and token ring data frames at 100 Mbits/sec. However, because it introduces a new MAC layer known as demand priority access--significantly different from Ethernet`s CSMA/CD approach--IEEE Working Group 802.12 was formed to handle its development. Among 100VG-AnyLAN features is a packet prioritization scheme intended to promote multimedia support and support of local area network diameters to 750 meters over Category 5 UTP, with a limit of four levels of cascaded hubs. 100Base-4T, on the other hand, permits a maximum of three linearly cascaded hubs on Category 3 UTP and supports a maximum network diameter of 220 meters.
Another Ethernet variation is isochronous Ethernet, which was originally introduced to support integrated voice and data traffic. This hybrid 16-Mbit/sec approach combines a 10Base-T packet network with a 6.144-Mbit/sec isochronous switched network, using a star-wired UTP cabling system. Recognizing the fundamental differences between the way voice and video traffic (over switched circuits) and data traffic are handled (in packets), isochronous Ethernet combines the data features of Ethernet with the best features of integrated services digital network for voice and video. The 6.144-Mbit/sec isochronous channel can be subdivided by inverse multiplexing into 96 separate 64-kilobit-per-second subchannels-- ISDN-B channels--which can then be used individually or together to handle voice and video traffic. The two traffic types can co-exist on the same network because they are in different formats. However, although this isochronous Ethernet approach is supported by National Semiconductor and IBM, it has generated little interest because it is viewed as a niche product.
Full-duplex Ethernet
The last approach to improving traditional Ethernet is full-duplex Ethernet. The basic concept is to remove the collision-detection function, permitting simultaneous transmit and receive. Although this approach can more than double the throughput of a typical 10Base-T link, some limitations exist.
Only those users with a multithreaded operating system that provides simultaneous input and output instruction streams--for example, OS/2 and Windows NT--can take complete advantage of full-duplex Ethernet. Users of other operating systems will only see a marginal improvement in performance. Even with these shortcomings, though, the full-duplex concept is becoming increasingly popular as a means of decreasing server congestion, with vendors looking at it for token ring, fiber distributed data interface and Fast Ethernet.
One local area network approach that has generated renewed interest is FDDI. This original 100-Mbit/ sec token-passing network architecture has mostly been used in high-bit-rate backbones supporting lower-speed horizontal local area network segments.
FDDI is a mature technology, with more than 100 vendors of products. Unfortunately, it has never really been a viable choice in the horizontal because of the prices of fiber media and electro-optics. However, with the recent development of inexpensive compact disk lasers and 780- to 850-nanometer laser-emitting diodes, there has been renewed interest in fiber-to-the-desk using FDDI.
The main advantage of FDDI is the immunity of fiber to electrical interference--both from external sources and crosstalk within the cabling system. Optical fiber lets FDDI use a relatively simple bit-coding approach--non-return to zero inverted--to support 500 stations on an FDDI ring and distances to 100 kilometers on singlemode fiber, while meeting a maximum bit-error rate of 2.5x10-10.
However, FDDI is basically a data- only architecture. Although there are upper bounds on packet delay, the delays are still variable. Also, FDDI packets can vary from 17 to 4500 bytes. This raises the question of whether FDDI can effectively support multimedia (voice and video) applications.
Taking advantage of all the advances in high-performance copper cabling system design, TP-PMD was developed to provide a low-cost alternative to FDDI. TP-PMD is a 100-Mbit/sec data-transmission protocol for shielded twisted-pair cable and Category 5 UTP copper cable.
Developed by FDDI Working Group X3T9.5 of the American National Standards Institute TP-PMD uses FDDI`s 4B/5B block encoding and a newly introduced three-level non-return to zero inverted bit-encoding algorithm known as MLT-3 to provide a 100-Mbit/sec data rate while still meeting Federal Communications Commission`s Class B emission requirements. Because the cost of copper electronics is substantially lower than equivalent 1300-nm FDDI electronics, TP-PMD is being seriously considered. Many vendors are now shipping standard-compliant UTP and STP hardware, proven to be very stable. This was demonstrated almost two years ago at the Interop West trade show, where 20 TP-PMD vendors demonstrated interoperability over a Category 5 UTP cabling system provided by three cable- and two connecting-hardware vendors.
Although FDDI and TP-PMD are primarily for data transmission, FDDI-II was an attempt by the ANSI Working Group X3T9.5 to support mixed-media traffic. Incorporating the original FDDI physical-layer characteristics, FDDI-II can dynamically partition its 100-Mbit/sec bandwidth into a token-passing local area network and multiple circuit-switched, isochronous wideband channels, each with a bandwidth of 6.144 Mbits/sec.
In turn, each wideband channel can be divided into subchannels with data rates in multiples of 8 kbits/sec. Individual FDDI-II stations can operate in either standard FDDI or hybrid mode. However, no mix of FDDI and FDDI-II stations is allowed on the same ring. While this implementation may see some niche use for multimedia networking, it is too expensive to be considered a general-purpose desktop solution.
FDDI Follow On LAN, another proposal developed within the ANSI working group, was designed as a high-speed backbone to support multiple FDDI and FDDI-II local area networks. This hybrid implementation, which is similar to isochronous Ethernet, is composed of a packet network for data and an isochronous network for time-sensitive voice and video traffic. The major advantages of this alternative are specifications that allow for ATM switching support and a scalable architecture that supports data rates to 2.4 gigabits per second. However, it is very expensive as a desktop solution, and there has been little interest in its further development.
The switching alternative
Not all recent approaches to this issue involve the introduction of new local area network architectures. Another approach that is gaining support is local area network switching, which involves replacing conventional local area network hubs with switched hubs that establish dedicated links to selected network nodes. With this approach, instead of continually resegmenting a local area network when its saturation point is reached, servers, power users or entire network segments can be provided with dedicated ports on the hub.
Local area network switching products are technology extenders; they work with existing cabling systems and adapter cards but they aren`t scalable. Although they can be very fast--because the switching is done in hardware--actual network performance depends on the network configuration.
Switching technology is not limited to any one network type. Ethernet products are already available, and both token ring and FDDI products are also on a limited basis. In fact, some products even provide multiple local area network-protocol support via individual port modules on the switch. However, with these products, communications between different network types must be via a router or bridge.
In the more sophisticated high-end products, the aggregate switch bandwidth is greater than 10 gigabits per second, and some products can support more than 150 individual local area network segments. Switching products provide an economical way to expand network capabilities while still preserving the investment in adapter cards.
Fibre Channel
One fiber implementation that is starting to gain renewed support, particularly in high-end applications, is Fibre Channel. This network was initially developed in ANSI Working Group X3T9.3 in 1988 as a high-performance serial link for data exchange between workstations, mainframes, storage systems and super computers. It has since been recognized as a potential solution for any application that requires sustained, high-speed asynchronous communication, including medical imaging and visualization, file transfer and mass storage.
Now being developed by ANSI`s X3T11 committee, Fibre Channel is supported by the Fibre Channel Association, a group of more than 15 vendors. With revision 4.1 of the physical-layer specification published in 1994, the first Fibre Channel products--fast disk-array subsystems--are starting to appear. It is expected that the first local area network products will come to market shortly.
The benefits of Fibre Channel stem from a hardware-intensive switched technology that addresses voice, data and video applications. This scalable architecture supports data rates from 133 Mbits/sec to 1.0625 Gbits/sec and addresses the three lowest layer services of the open system interconnection model--physical cabling, data link and media access, with mappings to upper-layer protocols.
Originally defined as a fiber-only architecture, Fibre Channel for bit rates of less than 100 Mbits/sec has been extended to include both singlemode and multimode fiber at 780 and 1300 nm, coaxial cable and shielded twisted-pair cable. The distances supported are in the kilometer range for fiber; for copper, they are limited to 100 meters. The 8B/10B block-encoding algorithm used ensures good error detection and correction, and because the algorithms can be implemented in silicon, performance is enhanced at low cost.
Because of its multimedia capabilities and other advantages, Fibre Channel may compete to some extent with ATM as a horizontal or campus-backbone technology. However, ATM has advantages in wide area networks, and Fibre Channel itself will use the ATM network for wide-area connectivity.
With all of the recent attention given to ATM, you might think that it was a radical new technology. In fact, it is based on a cell-switching concept that was developed in the late 1960s at AT&T Bell Laboratories. There is renewed interest in the technology mainly because of the increased need for data-only networks to carry mixed-media traffic.
A packet network is best suited to the bursty, time-insensitive nature of data traffic, while a time-division multiplexed network is best for voice and video traffic. However, either network becomes inefficient when trying to accommodate traffic of the other type. That is the reason for ATM`s current popularity: It provides the best available compromise for networks that must carry both data and isochronous traffic.
Scalable architecture
Asynchronous transfer mode is a high-performance, cell-switching network architecture with fixed-length 53-byte cells--48 for data and 5 for overhead. Delivering high bandwidth and the low, predictable latency needed to address real-time communications, ATM is a scalable architecture that supports data rates from 25 Mbits/sec to 2.488 Gbits/sec. In fact, it is the only technology that is scalable from local to wide area networks on a global basis.
ATM specifications are being driven by the ATM Forum, a consortium of more than 600 members. The 155-Mbit/sec UTP specification was approved in August 1994, while the 51.84-Mbit/sec UTP specification has been forwarded to the Forum for approval. ATM at 155 Mbits/sec over UTP is defined for 100-meter Category 5 horizontal links using products that are in compliance with Telecommunications Industry Association/Electronic Industries Association standard 568A.
ATM is similar to TP-PMD in that it is a two-pair approach. However, unlike 100-Mbit/sec TP-PMD--which utilizes MLT-3 bit encoding to ensure a low fundamental frequency and scrambling to help limit emissions spikes--ATM uses non-return to zero bit encoding and does not use scrambling. ATM at 51.84 Mbits/sec is currently defined as a two-pair approach on both Category 3 and 5 UTP links. Subrate transmission is also defined at 25.92 and 12.96 Mbits/sec. Maximum supported distances are dictated by bit rate and medium. The encoding methodology is 16-CAP, one of the algorithms developed by AT&T Paradyne; scrambling is required.
ATM has the potential to provide significant advantages in local and wide area environments. However, several issues remain unresolved, including local area network emulation. Communications between existing local area networks and ATM is critical, especially in the short term. This is because in the last four years, U.S. corporations have spent more than $10 billion on local area network electronics, most of which is incompatible with ATM.
Local area network protocols and applications cannot be supported over ATM without modification. Emulation promises to change that by defining how an ATM network can emulate enough of the MAC protocol of existing 802.X networks to allow higher-layer applications to be used unmodified. Emulation specifications now being developed will utilize the new "available bit-rate" service class to support legacy local area network traffic.
Local area network emulation will also address the concern about broadcast and multicast support on ATM. The ability to transmit one-to-many messages is not supported cleanly by ATM. Existing approaches put the onus on the ATM switch to make multiple copies of an input message and distribute it to the appropriate output ports on the switch.
At this time, it is agreed that ATM will not perform well for isochronous communications at the low bandwidths typically required for current desktop applications. This is because of ATM`s fixed-cell nature; the minimum overhead for an ATM cell is almost 10%, or 5 bytes out of 53. If the data to be carried does not fit exactly into the 48-byte packets defined for ATM, the overhead can actually be in the range of 20% to 50%, because unused packet space translates directly into overhead. When the effective bit rates get very high, the overhead issue becomes negligible.
In addition, as a result of the publicity surrounding ATM, every major local and wide area network vendor has announced an ATM product or a plan for one. However, because the private network-network interface specification that describes the interface between private ATM switches has not been completed, many vendors are using proprietary network-network interfaces. This raises the issue of interoperability between different vendors` products.
Most of the early sales projections for ATM have been optimistic. However, it is doubtful that widespread use of ATM in local area networks will occur before 1998 or 1999--and certainly not before some these issues are resolved. A "wait-and-see" approach is illustrated by the results of a 1994 Sage Network Research survey of 256 network managers at U.S. companies. The survey indicated that although more than 50% of the respondents will be evaluating ATM as a local area network technology, just as many will be looking either at TP-PMD or FDDI; even more will be looking at Fast Ethernet. Another 1994 survey by Intelliquest showed that the three most popular near-term local area network alternatives for the next year will be Fast Ethernet, full-duplex Ethernet and switched Ethernet, respectively.
Steve Bartolutti is a member of the technical staff at AT&T Network Systems, Norcross, GA.