Electronics Technology Assessment

The Interim Project continues to be implemented and is currently running ahead of schedule and meeting or exceeding the demands of the users it was intended to serve. During this period of time, an exhaustive evaluation of electronic alternatives has been conducted. This began with a "Think Tank" in which Information Technology invited representatives from Stanford University, UC Berkeley, UC Irvine, UC San Diego, and UC Santa Cruz to discuss technologies and vendor options. The conclusion of the Think Tank was that two viable options now exist rather than one: Fast Ethernet and ATM. The Think Tank members further concluded that because of a loss of vendor support for aging UCDNet equipment and increasing demand on the network, it was imperative to come to a quick conclusion on the appropriate technology and to quickly begin implementation. A draft document developed by the Think Tank members entitled "Network 21 Architecture" discusses the technology options in detail.

A more detailed analysis has been conducted over the last month to determine which of the technologies recommended by the Think Tank should be deployed in the final stage of Network 21. This report is a summary of the detailed analysis.

Timing of the Decision

As observed by the Think Tank group, the current interim network will last 12-18 months. Its life expectancy is limited by campus load growth, discontinued support on major portions of the existing UCDNet equipment and technology, and the lack of the final stage monitoring systems at the NOC to proactively manage a network of such expanded size. Given the length of time to prepare bid specifications, bid the electronics, perform interoperability and performance tests, and deploy the electronics, it is necessary to make a technology decision and begin implementation within the next two weeks.

Evaluation Criteria

The detailed analysis examined a total of twenty-two alternative configurations representative of the two technologies. A two step process was used to evaluate configurations that would be suitable for the UC Davis campus networking environment. The first step was to screen the options and to eliminate those that did not meet the set of screening criteria. The following step was to apply a set of ranking criteria in order to determine the most viable options. Many of these criteria are expanded upon in the Think Tank document.

Screening Criteria The analysis included performance objectives, cost, background vendor interviews, visits to several major technology vendors, and discussions with large-scale users of both technologies around the country. Upon completion of this analysis, two general screening criteria were used to narrow the field of options under consideration:

Minimum Acceptable Functionality

For the purposes of this evaluation, functionality was defined as meeting the following criteria:

Must support significantly more than 4 MAC addresses per NAM. An option that supports only 4 MAC addresses would restrict departmental flexibility to control cost through the use of fan out devices such as dumb hubs.
Must provide more than 1 Gigabit of effective backbone bandwidth
Must have the ability to route 2 million packets per second to and from VLANs/Subnets
Must enable the use of VLANs
Must provide a port concentration ratio of no greater than 2.4.
Must be currently commercially available and have comparable reference sites.

Cost

The maximum implementation cost to remain within the $23 million Network 21 budget cap.

Ranking Criteria The technology alternatives were assessed based upon several criteria. These criteria were ranked on a 1-to-5 scale with 1 being the worst, 3 being "average", and 5 being the best. For most items "average" can be considered to be representative of the original project plan. Criteria used for the ranking are as follows:

Costs, Initial: Estimated Initial installation cost including electronics, tax, shipping, with University Installation and Project Management.

Costs, Operations and Maintenance: Operations and Maintenance costs are affected by cost of maintenance contracts, difficulty of troubleshooting due to complexity and available tools, cost of spares, staffing levels and training costs.

Costs, Replacement: Reflects the amount of equipment that would be expected to be traded out during the next round of upgrades/life cycle, combined with the relative life cycle length. 1 roughly represents 3 years and 25 percent of the equipment would be retained, 5 roughly represents 4.7 years and 75 percent of the equipment would be retained.

Bandwidth, Initial: A gross quantification of backbone, uplink bandwidth, and concentration ratio. 1 represents 1.2 gigabits of effective backbone bandwidth, 100 megabits per uplink, a concentration ratio of 2.4, and the inability to load-balance in the backbone. 5 represents 1.4 gigabits of effective aggregate backbone bandwidth, 155 megabits per uplink, a concentration ratio of 1.55 and the ability to load-balance. It should be noted that aggregate backbone bandwidth is only an indicator of network performance, and that uplink speed and concentration ratio have as large or larger effects on performance as does aggregate backbone bandwidth.

Bandwidth, Scaleability: Four factors go into this criteria: (1) the ability to add or use additional links or higher link speeds; (2) The ability to increase speed incremental amounts without interrupting the operation of the network; (3) The ability to allow existing underutilized paths to be used to "load balance" and thus increase effective bandwidth; (4) The ease with which routing throughput be augmented. If an alternative can "load-share", have parallel links that will actually be used, have bandwidth augmentation installed on the fly without disruption, and have routing capacity easily added, it was awarded a 5. Each issue not addressed cost one point.

Bandwidth, Upgrade Path: A very speculative number assigned based upon how much of the existing installed base of electronics would require replacement in order to move to the next generation of technology. Roughly speaking, 25 per cent received a 5, 100 per cent received a 1.

Functionality, VLANs: Viewed as a very basic assessment of whether VLANs would help with our geographic diversity and frequent moves. All of the alternative finalists satisfy this requirement.

Functionality, Portability: This is a relative ranking of the complexity to the user and to the support organizations to empower "portable computing" which is defined as the ability to unplug a computer from one location and move to another and receive exactly the same functionality. If the alternative did this with a minimum of set-up, a minimum of administrative load, and smooth authentication, a 5 was awarded. A 1 indicates that it is not possible. Intermediate values indicate relative functionality, complexity, and cost.

Operability, Local Servicing: Three factors were used to assess the ability to perform local servicing: (1) quality of tools available; (2) the complexity of the task, and (3) the availability of spares and repair parts.

Operability, Maintainability: Three factors were considered: (1) resiliency of the system; (2) stability of the system; (3) simplicity of the system. To some degree, these factors can counterbalance.

Operability, NOC Operations: The primary issue with operability is complexity. There is some trade-off between training and position levels required and the cost and our ability to retain staff. A very complex system could require NOC staffing of one programmer-analyst 4, three programmer-analyst 3, and one programmer analyst 1, each requiring 4 to 5 weeks of training per year because of a multi-vendor environment. This would receive a rating of 1. A much simpler system might require only one programmer-analyst 4, one programmer analyst 3, and two CRS two's, each requiring two weeks of training per year. Such staff have become expensive and hard to retain because the systems for which they are trained are widely deployed in hundreds of smaller corporate environments all over the Bay Area where demand and pay scale is much higher.

Risks, Financial: Assessing financial risk from a technical perspective addresses such issues as past vendor performance in providing cost effective upgrade paths, long term vendor support of product lines, the vendor's length of time in the market, and the vendor's perceived consistency and stability of vision and direction. A 5 would go to a solution from a company that has been in the marketplace for over 10 years, the solution is central to the company's core competency and vision, the company hasn't abandoned product lines unexpectedly, the company's stated direction is and has been consistent, and the company is financially stable. A 1 would go to a solution from a company that has no real committed vision, that has abandoned product lines, that has changed direction at the expense of their installed base, or has been in the market so short a time as to not be able to assess these factors.

Risks, Functionality: The two main ranking criteria are the potential for advancement in functionality of the actual hardware under consideration, and the vendor's history for providing upgrades within the current product line rather than introducing new products for each incremental increase in functionality. A vendor that uses state of the art hardware at each stage can be a risk by pursuing an advanced design path that precludes the potential for future feature sets. A vendor that uses more flexible tried and true hardware solutions can be constrained in the performance area of functionality. A 5 would be awarded to technology just behind the bleeding edge due to having the highest performance, most advanced features, and least risk of a constrained future. A 1 would be awarded to a product suite consisting of sufficiently old technology as to be constrained by performance and capacity to support mainstream features, or is marketing a hardware-based proprietary solution that will soon be abandoned in lieu of standards-based hardware.

Risks, Operability: The dominant criteria for operability is lost productivity due to downtime. Complexity of a system factors into this as it opens the door to accidental configuration errors. These errors typically effect small segments of the user community. Heterogeneous spares induces another level of complexity that makes service restoration more difficult. System resiliency, however, is by far the dominant factor. For instance, a highly resilient system would not adversely affect the functionality anywhere other than where the single failure occurred. A value of 5 would go to a solution that has a minimum number of unique spares, has redundancy built in, has robust fault location tools, and affects service to only those users directly attached to a failed element. A value of 1 would go to a system that requires maintaining spares of multiple versions of a particular element that also subjects all users to service loss as a result of a single element failure.

Risks, Strategic: The dominant criteria for Strategic risk is the extent to which decisions made at this point would curtail options for the next technology cycle. A 5 would be assigned to having no constraints imposed whatsoever, i.e., an entirely standards-based environment with full interoperability and no constraints based upon platform dependencies. A 1 would be an entirely proprietary solution that could only be upgraded within a single vendor's product line.

Developing Options

The Two Technologies - ATM vs Fast Ethernet

The Think Tank had explored six options that the campus has in addressing its networking needs and subsequently narrowed the consideration to two based upon business drivers and suitability for the campus. Those two consisted of an ATM backbone with Ethernet to the desktop, and Ethernet and Fast Ethernet throughout.

ATM has been, for the last two years, the backbone technology of choice in networks of our size. FDDI has been declining rapidly in popularity due to cost and latency issues, and Fast Ethernet is just now coming down in price to the point at which it can be considered. There is also much effort going into developing other aspects of Fast Ethernet to increase its suitability in a large campus backbone. While much press attention is being paid to the prospect of Fast Ethernet backbones, ATM is currently being chosen by a factor of 2 to 1 in the campus environment. A qualitative discussion of the functionality of each technology is presented below or one may reference the Technology Features Matrix (Table 1) that was developed to compare the two technologies..

The networking environment at UC Davis is evolving rapidly into one in which both backbone and local bandwidth requirements are escalating rapidly and geographically separated groups of users need to be associated into logically contiguous LAN environments. This strongly suggests the need for Virtual LANs and easily scaled bandwidth.

Virtual LANs provide a practical solution to the geographic challenges faced by the bulk of the departments on campus. Virtual LANs may be implemented now with a proprietary solution over a backbone of fast ethernet technology. This will come at a future price of wide scale hardware replacement for the sake of achieving a supportable standards-based network. Virtual LANs may also be implemented now utilizing standards-based solutions using ATM across the backbone and to the wiring closet.

Bandwidth requirements can be met in the short term by implementing 100 megabit ethernet in the backbone, and then replacing core backbone links with gigabit ethernet once products for this are available.

Bandwidth requirements can also be met by implementing ATM at 155 megabits, adding load-sharing parallel links in the mid term, and then replacing core backbone links with 622 megabits and then further to higher ATM data rates. Standards and products are available now for this.

Multimedia applications can be supported with either an ATM backbone or a backbone composed of Ethernet variants. In the future, both ATM and Ethernet variants will be able to provide more uniform performance for multimedia applications respectively via QoS on emulated LANs or via priority queuing over Ethernet.

While there are evolving standards that allow an IP based application to request a Class of Service from the network, there are no standards for translating those requests into actual performance guarantees, and so Ethernet variants hold no promise for rigorous support of constant bit rate applications such as broadcast or video-conference quality video or dial-tone quality voice. There are standards now for ATM that provide constant bit rate service for broadcast and video-conference quality video and dial-tone quality voice to the desktop; however, additional equipment and software is required.

It is impossible to predict which technology will be dominant in the market place 2-3 years from now. ATM has won the battle for the WAN (Wide Area Network), and Ethernet has won in the LAN (Local Area Network or "desktop"), but it is likely to be 2-3 years before there is a clear indication of whether Fast Ethernet and Gigabit Ethernet will displace ATM as the technology of choice for campus backbones. The current battle is between ATM (at 155 megabits) and Fast Ethernet (at 100 megabits). It is expected that this battle will change to be one between ATM (at 622 megabits or 2488 megabits) and Gigabit Ethernet (at 1000 megabits) within the next 12 to 18 months. ATM is currently outselling Ethernet for backbones and is given the strategic advantage by several consulting firms. While ATM appears to be a superior technology, superior technologies do not always attain a dominant market share partly because consumer behavior is to avoid change and can largely affect the which products will dominate. In either case, the potential for minimizing life cycle costs must be taken into account.

Developing Specific Options

In order to compare the two technologies equitably, they were expanded into 22 distinct technology and design alternatives that consisted of the leading vendors' equipment applied to each technology and to the UC Davis environment. Eight vendors were pursued, most of which had both ATM and Fast Ethernet solutions, two which had only Fast Ethernet, and one which had only ATM.

In order for the technologies to be evaluated in the context of the UC Davis campus networking environment, a model was constructed of the intended campus network based upon the number of NAMs to be served, the scaling of the closets that serve those NAMs, and the topology of the fiber infrastructure.

In developing the design alternatives, the major network vendors in the campus backbone marketplace were interviewed. The leading vendors were identified via experience, reference, trade publications, and an informal market survey conducted at Interop. This list was confirmed via Gartner Group, a leading national technology consulting firm. Gartner Group was also asked to assess non-technical aspects of the vendors considered to assure that a mistake would not be made by perhaps selecting good technology from a poor business partner.

Since the major vendors tend to be somewhat conservative in their product offerings, two very young companies were also included. One of the young companies has already been acquired by one of the major vendors.

Each company was interviewed at least two times. The first interview was intended to assess the company's vision, target marketplace, and technologies futures. This level of assessment requires executing non-disclosure agreements with those companies. The second and subsequent interviews were targeted at actual design solutions and were taken to the detail of mapping the company's product line to our infrastructure and determining functional caveats and limitations.

Evaluation of Options

The screening criteria was applied to the 22 technology models. This resulted in the elimination of 7 models either because the vendor/technology alternatives did not fit the UC Davis campus environment owing to scaling issues (i.e., the equipment wasn't designed for a network of this size) or because their equipment was not actually shipping yet and could therefore not be assessed via a reference for a working installation.

The next screening criteria applied was the greater than 4 MAC address per port test. Eleven of the models did not pass this criterion. This left four remaining options.

The model was then used to estimate the installed cost of the remaining four alternatives. This eliminated two alternatives.

The remaining alternatives consisted of one Fast Ethernet and one ATM solution as referenced in the Summary Evaluation Table (Table 2).

Having completed the screening criteria, the ranking criteria was then applied to evaluate the remaining viable Fast Ethernet and ATM alternative. Scores were provided for each of the ranking criteria.

Preliminary Conclusion

There are two viable options for the Oversight Committee to consider - One which is Fast Ethernet, the other which is ATM. These options have somewhat different cost non-cost performance.

Glossary

acquired: Bought by another and larger company. Many large technology suppliers have found that it is less risky to buy the best of a particular new technology by "acquiring" a small and typically venture capital start-up company than to invest in in-house research and development.
aggregate bandwidth: The sum of the bandwidth available across a backbone, adjusted for shared links and multiple-hop paths.
authentication: The process by which an end user is identified both for admission control to a secured environment and for resource utilization tracking.
ATM (Asynchronous Transfer Mode): A connection-oriented network technology that uses small, fixed-sized cells at layer two. ATM has the potential advantage of being able to support voice, video, and data with a single underlying technology. ATM is the technology of preference in the wide area.
backbone: A high-speed network that connects several locations. The part of a network used as the primary path for transporting traffic between network segments.
bandwidth: The capacity and speed of a network, usually measured in bits per second. Network systems need higher bandwidth for audio or video than for e-mail or other services. Two categories of bandwidth are: broadband, which is faster and is used for complex telecommunications, and narrowband, which is the slower form and is used for voice and fax communications.
bleeding edge: Technology or hardware that is new and as yet unproved, often to the point of being marginally reliable and difficult to operate and maintain.
cell: In an ATM environment, layer three packets are divided up into 48 byte pieces and inserted into cells for device to device transmission. The cell headers contain only a "connection identifier" that is mapped to source and destination addresses at each transmitting or receiving device.
Class of Service: A way of expressing in simplistic terms the performance expectations associated with a particular application or session. This is used mostly in the context of the TCP/IP (layer three) environment. An analogy can be drawn to air travel: First class, coach, and stand-by are analogous to Classes of Service; whereas scheduling, length of lines, speed of a flight, width of seats, quality of meals, etc. are all analogous to Quality of Service issues.
concentration ratio (c/r): The amount of desktop bandwidth that shares uplink bandwidth expressed as a ratio. Lower numbers are better. This ratio affects performance now, and will affect performance more as desktops become more powerful and the Web gets used more for common applications.
constant bit rate service

broadcast quality video
Video that has similar or better quality than that which can be attained using a VHS video recorder.

video-conference quality video
Somewhat less quality than broadcast quality, slightly less detail, somewhat stuttered motion due to fewer frames per second. A bit like watching an 8 millimeter movie. Most commercial video conferencing systems meet the same standard for quality.

dial-tone quality voice
Voice transmission that has the same quality as a local telephone call. Cell phones do not meet this level of quality.
dumb hub: A hub that has typically between four and eight ports, sometimes more, used in a office work area to connect additional work stations to a single wall jack connection. Often times referred to as a fan-out device. Does not have status report-back intelligence to a central network management system.
edge device: A physical device which is used to forward packets between desktop network interfaces (such as Ethernet or Token Ring) and backbone transport facilities (such as ATM or trunked gigabit ethernet). This forwarding is based on either data-link or network layer information. Edge Devices typically reside in wiring closets (IDFs) and are the meeting point for the horizontal wiring (copper) and the riser cables (fiber).
electronics: The physical devices that make up the network, which include gateways, routers, switches, and edge devices.
emulated LANs: ELAN, is a logical LAN as a result of LAN Emulation (LANE). Essentially, it is a LAN emulated on top of ATM. This mechanism is used to implement Virtual LANs in an ATM environment.
Ethernet variants

Ethernet
A legacy LAN protocol, known as the IEEE 802.3 standard, was originally designed to operate up to 10 Mbps. This popular Local Area Network technology was invented at the Xerox Corporation. An Ethernet consists of a cable to which computers are attached. Each computer needs hardware known as an interface board to connect the computer to the network. Various 10 Mbps Ethernet systems have emerged: 10 Base-2, 10 Base-5, and 10 Base-T.

Fast Ethernet
One of the more recent evolutions of Ethernet, this high-performance LAN technology enables data transmission at 100 megabits per second.

Gigabit Ethernet
The most recent evolution of Ethernet, this high-performance LAN technology enables data transmission at gigabits per second (Gbps) speeds. The standard is not expected to be complete until the end of the first quarter of 1998. Some vendors are offering pre-standards products.
fan-out device: Otherwise known as a dumb hub; see definition for dumb hub.
FDDI: Fiber Distributed Data Interface. An ANSI defined standard for implementing a high-speed (100 Mbps) LAN that spans up to 100 meters and is based on a dual fiber optics token-ring topology.
frame: In an Ethernet environment, higher layer network protocols are encapsulated in frames for device to device transmission. Each frame contains a packet surrounded by MAC addresses for the source and destination of the message.
gigabits: A billion bits per second or a thousand million bits per second.
Interop: A most highly regarded conference addressing layer two and layer three computer networking issues.
IP based application: Applications dependent on Internet protocols for inter-computer communications.
LAN (Local Area Network): Any combination of more than two computers connected together to share files or printers.
large-scale users: Typically campus environments supporting over ten thousand end-stations.
latency: The delay in propagating a message from one end station to another through a network. This delay is only vaguely related to the transmission speed or bandwidth, and is in fact introduced by network devices along the path. This is done proactively.
load balance: The ability for two paths from one point to another in a network to both be used to their peak efficiency. This increases efficiency by forcing under-utilized paths to divert traffic from over-utilized paths. This is typically passive and reactive.
load-sharing: The ability for two paths from one point to another in a network to both be used to some degree. This increases efficiency by allowing under-utilized paths to compensate for over-utilized paths.
local servicing: Maintenance and repair that is performed by in-house support personnel as opposed to outsourced support. The advantage is typically more rapid response and restoration due to proximity and intimate knowledge of the local environment.
MAC address: Media Access Control address, otherwise known as the hardware ethernet address, normally expressed as a hexadecimal number.
MACs per Port: Various electronics alternatives are able to support different numbers of desktop devices per NAM via fan-out devices. These numbers fall into three categories: tightly constrained (4 per NAM), somewhat constrained (average of 20 per NAM or 512 per uplink), and broadly constrained (average of over 40 per NAM or over 1024 per uplink). For the broadly constrained alternatives, the actual number wasn't annotated. The larger the number of MAC addresses supported, the higher the cost of the edge device.
megabit: One million bits per second.
multimedia applications: Applications that provide an integrated way of presenting to the user a combination of different kinds of information such as text, data, images, video, audio, and graphics.
NAM (Network Access Module): Commonly known as a wall jack or work area outlet.
Network: A group of computers, switches and connections that communicates information between users.
NOC (Network Operations Center): This facility provides monitoring, performance analysis, and repair coordination for the campus-wide network. The NOC for the UC Davis campus is located in the Telcom Building.
packet: The basic message unit for data transfer between computers over a network. The various ISO layer three network protocols use different packet sizes. TCP/IP has a size range of between 64 bytes and 1514 bytes. Layer three packets are encapsulated in various layer two units called frames or cells.
parallel links: Two or more transmission paths that either may or may not be able to be used simultaneously for either load balancing or load sharing.
plant infrastructure: In general, anything permanently installed to make a building usable such as electricity and communications wiring, plumbing, and air conditioning. In the context of Network 21: all the facilities such as the fiber, copper, inside and outside conduit, risers, closets, cabinets, backboards, patch panels, and punch blocks.
priority queuing: A mechanism for which a standard is being developed that will provide for preferential treatment of certain data transmissions. This methodology will provide no guaranteed level of service, but will provide the preferred traffic with a "best effort" delivery. The standard will apply to Ethernet based transmission.
proprietary: Something that is used, produced, or marketed under exclusive legal rights of a company. Not standards-based.
QoS (Quality of Service): A mechanism for guaranteeing performance in delivery of specified data transmissions across a network. This enables sufficient control over performance to allow for constant bit rate applications to work well across a data network. This mechanism is available only in an ATM environment.
redundancy: Serving as a duplicate for preventing failure of an entire system upon failure of a single component.
resiliency: The ability to recover from or operate uninterrupted upon a change.
scaling: Adjusting to a larger base without repercussions.
subnet: A networking scheme that divides a single network into smaller logical networks to simplify routing.
standards-based: Conforming to models established by professional and governmental organizations to establish uniformity and compatibility in products.
UCDNet (University of California, Davis Network): The UC Davis campus high speed network, operated and maintained by Information Technology, Communications Resources.
uplink: The communications link that connects an edge device in a local wiring closet to the network backbone. This link represents the first potential bottleneck to performance.
VLAN (Virtual LAN): A networking environment where users on physically independent LANs are interconnected in such a way that it appears as if they are on the same LAN (workgroup). This means that membership in a LAN environment is no longer constrained by geography. Membership to a Virtual LAN is defined administratively independent of the physical network topology and requires centralized administration. A Virtual LAN segment is a unique broadcast domain.
WAN (Wide Area Network): A physical network that spans large geographic distances and transfers data between multiple LANs. Also called long-haul networks.
wiring closet: An enclosed, secure, clean and environmentally controlled and conditioned space for housing communications equipment, cable terminations, and cross-connects. This closet is the recognized cross-connect between the backbone cabling or riser cabling and the horizontal cabling.

[UC Davis][Information Technology][Communications Resources][Network 21]
Administration by: WebKeeper - Editing corrections to: Erin Magers
Last modified 10/14/97