ATM: Reality or Pipe Dream Copyright CAUSE 1994. This paper was presented at the 1994 CAUSE Annual Conference held in Orlando, FL, November 29- December 2, and is part of the conference proceedings published by CAUSE. Permission to copy or disseminate all or part of this material is granted provided that the copies are not made or distributed for commercial advantage, that the CAUSE copyright notice and the title and authors of the publication and its date appear, and that notice is given that copying is by permission of CAUSE, the association for managing and using information resources in higher education. To copy or disseminate otherwise, or to republish in any form, requires written permission from CAUSE. For further information: CAUSE, 4840 Pearl East Circle, Suite 302E, Boulder, CO 80301; 303-449-4430; e-mail info@cause.colorado.edu ATM: REALITY OR PIPE DREAM Douglas Gale Guy Jones University of Nebraska-Lincoln Lincoln, Nebraska Martin Dubetz Washington University St. Louis, Missouri Abstract Since the early 1980s, campus networks have generally been based upon shared backbones. In this paradigm, traffic from individual users is aggregated on backbone networks and each user gets a share of the backbone bandwidth. That paradigm is rapidly becoming inadequate to meet growing user demands. This paper describes the strategies developed at Washington University and the University of Nebraska-Lincoln to transition our campus networks to Asynchronous Transfer Mode or ATM technology. Paradigm Shift Since the early 1980s, campus networks have generally been based upon shared backbones. In this paradigm, traffic from individual users is aggregated on backbone networks and each user gets a share of the backbone bandwidth. That paradigm is rapidly becoming inadequate to meet growing user demands. While the original motivation to move towards higher speed "broadband" networks was new applications such as multimedia and client server, those applications have happened more slowly than anticipated and the current motivation for broadband networks is traffic aggregation and LAN interconnections. While it is difficult to quantify the growth in peak bandwidth demand or even average bandwidth utilization on a campus network or local area network, there exists good data on the growth of average bandwidth utilization on the NSFNET, the primary component of the Internet backbone. That growth is shown in Fig. 1. [FIGURE 1 NOT AVAILABLE IN ASCII TEXT VERSION] It is reasonable to assume that the grown of campus network traffic parallels the growth of the NSFNET since the NSFNET is the composite of the intercampus Internet traffic. The composition of that traffic is changing. In addition to the growth that is resulting from a larger user base, the network is being increasingly used to transfer multimedia information using software such as "Mosaic." This shift is illustrated in Fig. 2. [FIGURE 2 NOT AVAILABLE IN ASCII TEXT VERSION] It is clear from the data that the traditional uses of the Internet, file transfer (FTP), electronic mail (SNMP) and interactive logons (Telnet) are a decreasing portion of the utilization of the Internet. The growth in "Other TCP/UDP Services" can be attributed to multimedia applications such as "Gopher" and "Mosiac." Separate data indicates that Mosaic has been growing at 61% monthly since its inception. Not only do we feel that bandwidth requirements will continue to increase at their historical rate, we are also concerned that historical growth patterns may not adequately reflect the bandwidth requirements of interactive multimedia. In other words, the projections that follow may be far too conservative! Our campus experiences have indicated that multimedia creates a quantum jump in bandwidth requirements. In an attempt to project the future bandwidth requirements of campus networks, we have extrapolated the historical growth of the NSFNET through the remainder of this century. Our assumption will be that the growth of campus peak and average bandwidth requirements will parallel or exceed the historical growth of the NSFNET. That extrapolation is shown in Fig. 3. [FIGURE 3 NOT AVAILABLE IN ASCII TEXT VERSION] The data clearly indicates bandwidth increases of two orders of magnitude (a factor of 100) by the end of the decade. Many campus networks, including the University of Nebraska- Lincoln and Washington University, have already reached the point where shared bandwidth backbone architectures are no longer capable of meeting user demands. Bandwidth increases of two orders of magnitude cannot be accommodated with current architectures. Many campus networks will reach this point within the next few years. The most obvious solution is to increase the speed of the shared backbone. Unfortunately, the cost of increasing the speed of a network that spans hundreds of kilometers is not easily accomplished. Using current technology it is very expensive to increase the speed of shared backbone networks one order of magnitude. There are no technologies available or proposed that provide two orders of magnitude increase for wide-area shared backbones. Current technical solutions to this bandwidth problem all involve reducing the dependence on "shared backbones" and a migration towards dedicated bandwidth. Alternatives to the Current Paradigm There are a number of alternatives to the current shared backbone paradigm. By segmenting current routed backbone networks, we can effectively reduce the traffic on the shared backbone component. Ultimately, however, such techniques reach a point where more fundamental changes are necessary. Switched Ethernet. Switched ethernet utilizes a star wiring configuration to extend 10 Mbs links from a high speed central switch (several hundred Mbs) to distributed equipment. In a typical campus environment, these distributed units might be distributed routers and strategic computing resources. This strategy suffers from two limitations. The first is that it is not scalable. Ultimately the 10 Mbs (or 100 Mbs fast ethernet) link to the distributed units will be overwhelmed. The second is that ethernet is not isochronous. That limits the use of the technology to data and low grade video or voice. In our opinion, switched ethernet should be regarded as a transition strategy. Switched FDDI. Switched ethernet utilizes a star wiring configuration to extend 100 Mbs FDDI links from a high speed central switch to distributed equipment. In a typical campus environment, these distributed units might be distributed routers and strategic computing resources. This strategy suffers from two limitations. The first is that it is not scalable. Ultimately the 100 Mbs link to the distributed units will be overwhelmed. The second is that FDDI is not isochronous. Again, that limits the use of the technology to data and low grade video or voice. In our opinion, switched FDDI should be regarded as a transition strategy. Asynchronous Transfer Mode. Asynchronous Transfer Mode or ATM is a fundamentally different technology. It is connection oriented, whereas ethernet and token ring are connectionless. It is scalable in multiples of 51 Mbs through several Gigabits per second. It was designed to carry voice, video, and data traffic. It represents a unifying force in that it provides services to workstations, computers, networks, homes, video stations, and telephones and is supported by both the computer and telecommunications industry. Problems Associated With ATM If ATM is so great, why isn't it being adopted everywhere? First, it isn't ready yet. There are no large ATM networks in operation. Second, the standards for ATM are still evolving. In particular, the standards for LAN emulation have yet to be finalized. Institutions have a substantial investment in current LAN's. Any successful transition strategy must provide for operating existing networks over ATM. Third, standards for quality of service (QOS), which is necessary for providing video and voice services is still evolving. Fourth, standards are not yet in place for linking equipment from different vendors. Fifth, video and voice may not use ATM technology. Video and voice are both very mature technologies and are very cost effective. The cost savings offered by an integrated technology may not be sufficient to offset the efficiencies already developed in specialized voice and video transmission systems. Sixth, LANs may not convert to ATM. The availability of inexpensive higher speed LANs, such as 100 Mbs ethernet, will reduce the demand for a transition to ATM. Standards have been developed for isochronous ethernet (IEEE 802.9). Multicasting bandwidth reservation will be available through enhancements of traditional TCP/IP. Transition Strategies The fact remains that there are no alternative to ATM that will meet the projected bandwidth requirements by the end of the decade. Both Washington University and the University of Nebraska-Lincoln began a transition to ATM technology several years ago. The transition strategies developed at both institutions have three components. The first is to deploy a wiring plant capable of supporting both future ATM technology current non-ATM technology. The second is to deploy transition architectures that will meet current campus needs while ATM technology is maturing. The third is to deploy ATM test bed networks to develop expertise and experience. Wiring Plant Given unlimited resources every location would have available STP, UTP, Coax, single-mode fiber optics, and multi-mode fiber optics. Cost and physical constraints imposed by retro- fitting existing buildings requires a more modest approach. At Washington University the following set of guidelines have been adopted. UTP: Level 5 UTP is installed at all locations. Where possible, multiple sets of two pair cables are installed but 4 and 8 pairs cables are also used. All cables are terminated at level 5 using punch down blocks in the wiring closets. Level 3 punch downs can be used for phones. When practical, the enhanced level 5 cable is used (sometimes called 5+ or level 6). Cable Runs: A 90 meter rule is used from the punch down block to the jack on the wall. This leaves 10 meters of cabling for connections to the host and the hub. Multiple punch downs on a cable run are avoided requiring some home running of cable. Wiring Closets: Wiring closets now contain hubs and required ventilation and power. Sufficient space in required for the equipment and maintenance access. Additional Station Cabling: Coax (CATV) is installed in residence halls, lounges, conference rooms and classrooms. Mutli-mode fiber is also installed in classrooms and conference rooms. Where possible, the fiber is left unterminated to reduce costs, but in many cases the difficulty of post termination (short leads) requires the fiber be terminated during installation. If open conduits are available for future installation only the copper is installed. Vertical Wiring: Both multimode and single mode fiber are installed to every wiring closet. In some cases two or three floors will share a closet. Level 3 trunk cable is pulled between floors for telephone. A common fiber de-mark consist of 18 multimode and 6 single mode fibers. Campus Wiring: The campus is divided into sectors each with a major fiber hub and several minor fiber hubs. Sectors are interconnected with 96 multi-mode fibers and 48 single mode fibers. The resulting topology is a inter-connected multiple star network. From this topology rings and buses can be created if needed. The major hub room will contain substantial equipment and additional attention is given to space and environmental considerations. Development and Training: Installation of ATM capable networks requires close attention to level 5 installation standards. Early training of engineers and technicians is critical for a successful installation. Legacy Systems Support: In nearly all cases, existing asynchronous lines (terminals) and ethernet connections are re-established using the new stations wire. Telephone trunks were installed at level 3 to provide connectivity to centralized building hubs or traditional asynchronous data circuits including fire alarms, security, and physical plant controls. Existing fiber rings and ethernet circuits are re-established by appropriate jumpering of the fibers at the major and minor hubs. If possible baseband ethernet cabling is left installed, if not ethernet is moved to fiber in a star configuration. Transition Architectures The second component of our transition strategy is to deploy a transition network architecture appropriate to the demands from users for more bandwidth and greater network capabilities. Differences between campuses will result in different architectures. Washington University has deployed switched FDDI, switched ethernet, and collapsed backbones. The topology of this deployment, shown Fig. 4 on the next page. The network consists of inter-connected stars; the same topology will be used for ATM. Networking privacy, a feature of ATM technology, is currently provided to the user by deploying secure hubs. The new generation of these hubs simplify the management of a privacy enhanced network. [FIGURE 4 NOT AVAILABLE IN ASSCII TEXT VERSION] At the University of Nebraska-Lincoln, bandwidth requirements track several years behind those at Washington University and allow a different transition strategy. UNL plans to by-pass both switched ethernet and switched FDDI and deploy ATM switching technology; Washington University has used FDDI switching as an interim step. UNL feels that enhancements and further segmentation of its current network will allow the campus to "get by" until ATM technology matures sufficiently for its adoption in a production environment. ATM Test Beds ATM technology is new and both campuses have deployed ATM testbed networks to gain experience. UNL ATM Testbed: The UNL testbed is a joint project of the Computing Resource Center, the Department of Computer Science, and the University Library. The testbed is shown in Fig. 5. [FIGURE 5 NOT AVAILABLE IN ASCII TEXT VERSION] Washington University ATM Testbed: The Washington University ATM network is shown in Fig. 6 on the next page. The switch speeds range from OC3 (155Mb/s) to OC192 (10Gb/s). Most of the OC3 switches are deployed, others will be added as they become available. The network is designed to bring ATM to the desktop. The figure shows a variety of hosts that contain commercial ATM SONET interface cards and other hosts utilizing a device called a multimedia explorer. This device supports bi-directional broadcast quality NTSC video and CD quality stereo sound through external ports. There is also an ATM host interface. The multimedia explorer is used for multimedia applications development on the ATM network. [FIGURE 6 NOT AVAILABLE IN ASCII TEXT VERSION] Selling ATM - The End Users View As a scalable and secure technology, ATM solves many network engineering and management problems. End users, however, will be reluctant to pay more money to adopt this technology if they do not perceive that they are receiving new and innovative services. The argument that "we need to do this to keep your email running the way it always has" is not likely to generate support for additional network expenditures. At both Washington University and the University of Nebraska- Lincoln, we feel that the development of new applications such as multimedia and remote collaboration (teleconferencing, telemedicine, distance learning, etc.) is an essential part of any ATM networking plan. To complement ATM technology Washington University has identified applications in Neuroscience, Radiology, Biology and Hearing where the strength of the ATM technology can make a solid contribution. The networking research group in the Applied Research Laboratory in the School of Engineering is working with these researchers and several prototype applications are operational. At the University of Nebraska-Lincoln we have created "New Media Center" dedicated to supporting the development of multimedia applications in instruction. That facility includes two completely equipted classrooms as well as a staffed media development area.