Gigabit Ethernet-Round II
With ratification last June of the IEEE 802.3ab standard for gigabit-per-second transmission over copper cabling, network designers, cabling contractors, and end users face more choices than ever
With ratification last June of the IEEE 802.3ab standard for gigabit-per-second transmission over copper cabling, network designers, cabling contractors, and end users face more choices than ever. In our second annual "virtual roundtable," we tap into the expertise of industry gurus to help answer our readers' concerns about Gigabit Ethernet implementation.
Participants include Hugo Draye, market segment manager at Fluke Networks Div. (Everett, WA); Hank Dunnenberger, director of premises marketing at Siecor (Hickory, NC); Robert Jensen, technical services team leader for the Volition system at 3M Telecom Systems Div. (Austin, TX); Scott Stevens, director of technical services at CommScope (Claremont, NC); and Peter Williams, general manager, WireScope Operations, Agilent Technologies (Marlboro, MA).
The first "virtual roundtable" of experts appeared in the Gigabit Ethernet supplement to the February 1999 issue of Cabling Installation & Maintenance.
Q: Some end users are currently planning a cabling upgrade to accommodate a move from 10-megabit-per-second Ethernet to 100-Mbit/sec Fast Ethernet to the desk, with an eye toward Gigabit Ethernet (GbE) in the future. What infrastructure type would you recommend to accommodate Fast Ethernet today and GbE sometime down the road?
Draye: 100Base-TX (Fast Ethernet) will run just fine on a Category 5 installation. In fact, the requirements for the physical medium for 100Base-TX as delineated in IEEE 802.3u are far less demanding than the Category 5 channel specifications of the Telecommunications Industry Association and the Electronic Industries Alliance (TIA/EIA-Arlington, VA). From a practical and economical viewpoint, I would recommend that end users planning either a new installation or a major upgrade install at least an Enhanced Category 5 (5E) cabling system. The material cost is just marginally higher than a plain Category 5 installation, and the labor costs should remain about the same as a Category 5 installation.
If an end user is going to occupy the building for more than 10 years, it may be advisable to select Category 6 cable. I would continue to select Category 5E connecting hardware at this time and verify that the installation meets Category 5E performance with some amount of headroom-a minimum of 5 decibels. Pulling the cable is, by far, the more costly and disruptive activity. I am not going to recommend Category 6 connecting hardware until interoperability of Category 6 connecting hardware is assured. If Category 5E connecting hardware has been installed, it may have to be replaced. Recall that GbE should run with ease on a Category 5E channel. Therefore, the Category 5E connecting hardware will not have to be replaced until such time as the application in the horizontal (to the desk) demands more than the current specifications for GbE (1000Base-T).
Any current version of Category 6 connecting hardware may also have to be replaced if an application more powerful than GbE is to be supported in the horizontal. Vendors of network electronics are going to select one type of Category 6 socket or jack in their equipment. This selection may not be compatible at the Category 6 performance level with any of the current vintage systems. Who would want to deploy "hybrid" patch cables?
Dunnenberger: Some end users appear to enjoy the constant recabling cycle. We believe it is actually a management ploy, implemented through IT departments, to prevent employees from becoming too settled and complacent in their work areas. The disruption caused by the periodic recabling exercise keeps everyone on his toes. If this is, indeed, the case, the end user should buy generic Category 5 cable and trust that the standards and installer practices will hold up to support their 100-Mbit/sec needs. Then they can look forward to and start planning another recabling when they are ready to implement GbE!
For those end users who don't enjoy the hassle of recabling or are frustrated by their inability to quickly adopt new applications because their cabling is not up to snuff, it is time to carefully evaluate the available cable choices. With GbE capability in mind, there are three approaches:
- Look at the standards and deploy generic Category 5 and keep your fingers crossed.
- Listen to the hype and mitigate risk by buying Category 5E-but be prepared to open the wallet wide.
- If Option 2 is being considered, as seems to be the trend, then we recommend that the end user still deploy a combination of copper and fiber to each workstation.
Surprised to hear this from a fiber vendor? In fact, such a strategy is being deployed by a growing number of users because it is risk-free and surprisingly affordable. From a performance standpoint, the copper can handle voice and some low-speed data requirements. Category 5E will do the job up to 100 Mbits/sec if it is installed carefully. Fiber is a no-brainer for 100Base-FX and the emerging 100Base-SX products, as well as GbE. However, it is still questionable whether copper will be practical and robust enough over time to support gigabit-per-second speeds without signal degradation due to crosstalk, temperature fluctuations, mismatched patch cords, cable stresses, and similar problems.
From a cost standpoint, the equation is appealing. It is quite common for an end user to run three or four sheaths to each workstation. Often these sheaths all end up being the highest grade of copper to allow versatility to accommodate either voice or data. The recommendation is to replace one of these copper sheaths with a fiber cable with two or four fibers. Depending on the category of copper selected, the "copper and fiber" scenario is either slightly more or slightly less expensive than deploying all copper.
The main hurdle for using fiber-to-the-desk (FTTD) networks is the cost of the electronics, not the cost of the cabling. But as network speeds increase and with the advent of the small-form-factor fiber-optic connectors, this cost delta is narrowing quickly. For 100-Mbit/sec Fast Ethernet electronics, MT-RJ-equipped fiber switches are within 25% of the cost of RJ-45 copper ports. In summary, the "copper and fiber" recommendation enables the user to run today's lower-speed copper-based applications economically while providing a very affordable low-risk solution to tomorrow's high-speed applications over the fiber.
One final note: When you upgrade the horizontal segment of a network, also evaluate the need to upgrade the backbone infrastructure accordingly.
Jensen: I'd suggest that users install FTTD. Fiber is a medium that resists the effects of electromagnetic interference (EMI) and can be used to migrate networks. All that is needed is the proper electronics. Today, there are less-expensive solutions to both cabling and networking equipment.
Copper networks have their limitations, especially for migration. For instance, migrating from 10Base-T to 100Base-T2 and 100Base-T4 saw significant problems with delay skew in customers' embedded Category 5 cabling. Migration to higher-speed networks also has these problems. As an example, even though Category 5 is stated as meeting gigabit-per-second capability in the IEEE 802.3ab standard, all electronics vendors are specifying Category 5E.
In addition, copper solutions continue to evolve with different categories of cabling. Even though Category 5 was supposed to be the "do-all, end-all" of cabling, manufacturers are now specifying Category 5E, 6, 6A, and 7-even though there is no standard for most of these cabling types. The bottom line is that unless you want to replace your cabling when you replace your network, install fiber today.
Stevens: The choice of cabling infrastructure always depends on the balance between performance and cost. In an ideal world, where budgets were not a constraint, a FTTD solution would be the only choice. More realistically, users should consider three solutions and choose one based on their budgets:
- The optimum choice is to run high-performance multimode FTTD for data and a Category 5E unshielded twisted-pair (UTP) cable for voice. The multimode fiber should have a bandwidth above 500 MHz-km to support future 10-Gbit/sec solutions or other technology advances (e.g., full-motion videoconferencing, video streaming, etc.). While this solution would provide the best measure of futureproofing, the higher per-port cost for optical line cards and network interface cards (NICs) would be significantly more than a UTP solution. This solution then lends itself to those who have bigger budgets or those who need the greatest hedge against technology advances.
- Unfortunately, budget constraints are a reality. High-performance Category 6 systems that come with warranties will deliver the same performance as FTTD solutions at a lower cost. Category 6 cabling will support the migration path through 1 Gbit/sec and probably on to 2.5 Gbits/sec. However, at 10 Gbits/sec, UTP may finally run out of bandwidth, barring some significant development in digital signal processor (DSP) technology.
- A Category 5E system would be the low-end solution for those who are under more-severe budget limitations. The downside is that at gigabit speeds, the Category 5E cabling may not perform as well as Category 6. Current industry testing has shown that the performance level of UTP cabling used can directly impact the bit-error rate of the system. The variability in transceivers at gigabit speeds coupled with lower-performance UTP cabling can cause significant degradation of system throughput or latency. It's the old adage: "You get what you pay for."
Even though cost would be the reason to consider Category 5E, the total cost savings over a Category 6 solution will be minimal. First, most of the costs are in the electronics, so the percentage of savings from the cabling would be small. Second, at gigabit speeds, there may be two versions of transceivers with different DSPs to support high- and low-end cabling. The savings in cabling costs of Category 5E systems may be offset by the need for more-expensive digital signal processors.
Williams: There is a broad consensus that new copper cabling to the desk should be at least Category 5E if the user intends to ultimately support 1000Base-T. So our recommendation for most commercial office environments would be to install either Category 5E or Category 6 to the desk and install either high-grade multimode fiber or both multimode and singlemode fiber in building backbones.
Q: GbE is currently being deployed predominantly as a backbone protocol. When do you predict that GbE-to-the-desk will become common?
Draye: My crystal ball is cloudy on this issue. I estimate that less than 10% of users today are connected in the horizontal with a link that delivers more than 10 Mbits/sec. Many predictions have been made that more bandwidth-hungry applications are going to be deployed in the near future. I predict that it will take about five years for Fast Ethernet to reach more than 50% of end users and that GbE will be used in about 10% of the desk connections by 2008.
Dunnenberger: Judging from the various industry forums we have attended over the last year, I believe GbE-to-the-desk will likely become common in the next two to three years. Why? Because there is a constant shifting of the network bottleneck from desk hardware to computer software to switches to servers, and so on. All of these pieces are changing and improving very rapidly. As an industry, we have a difficult time predicting what we might need to be doing in the future to warrant a gigabit pipe, but we only have to stop and look back a few years to see how far we have come and how fast.
Another consideration is that you don't need the full gigabit potential to warrant its deployment. In fact, anything over the nominal 100 Mbits/sec will put you in the league for gigabit-to-the-desk. Finally, there will be incentives from network electronics manufacturers to encourage the shift to the gigabit level. Performance has always been a strong differentiator in the competitive network electronics market. This incentive will come in the form of attractive price steps to move upward to the newer technology.
Jensen: GbE-to-the-desk will take a few years. Then, all that bandwidth will be needed to support voice, video, and data-all at the same time. Imagine a conference call with users discussing a 3D model displayed on their PCs while able to see each other. This may seem far off, but it is reality in some R&D facilities, including universities. Applications may also be run directly from a server rather than from the user's local computer. Essentially, there could be "dumb" terminals that do smart things with all the processing in one spot: the server. This will need bandwidth, and the backbone is always the part of the system that needs added bandwidth first. One of the most bandwidth-intensive applications today is just backing up a server.
Stevens: In order for GbE to become commonplace at the desktop, three things need to occur:
- First, the development of a 10-GbE needs to be finalized so that the backbones can support multiple 1-Gbit/sec feeds. IEEE is in the early stages of a 10-GbE standard.
- Second, the per-port cost of GbE needs to fall. This will occur with volume over time and will occur with the release of copper-based GbE electronics.
- Third, there needs to be an application(s) requiring the tenfold bandwidth gain of GbE over 100-Mbit/sec Fast Ethernet.
Only a small number of users can even begin to use 1 Gbit of bandwidth. Yet, if history is an indication, we have always found a way to use bandwidth. Graphics- or video-intensive Internet services or full-motion videoconferencing may be one of those bandwidth-intensive applications. More than likely, it will be something we cannot even envision yet.
Williams: GbE-to-the-desk will probably not become common until the price of a 1000Base-T NIC drops to less than $75-which may only take two to three years. However, once the price of 1000Base-T hardware hits the point where the premium is modest compared to the cost of 100Base-T, adoption will be widespread and rapid. At that point, many users will implement GbE-to-the-desk, whether or not their employees truly need the extra bandwidth.
Q: Some in the industry predict that 1000Base-T will be the "nail in the coffin" for FTTD. What effect, if any, do you think ratification of IEEE 802.3ab will have on implementation of FTTD?
Draye: FTTD will be selected for other reasons than 1000Base-T applications, for example, security, immunity to EMI, and considerations of centralized wiring structures.
The economic issues of horizontal cabling include the cost of the electronics at either end of the link. It is anticipated that the cost of GbE interfaces for copper will initially be rather expensive. That is to say, the cost per port of switches and concentrators and the cost of NICs will initially be high. Typically, the cost of silicon chips drops over time as production volumes rise. But if the cost of these copper interfaces remains high over time, fiber may offer a very competitive or better value for the investment. I predict that in four to five years, the cost of a copper GbE installation in the horizontal will still be lower than that of a fiber installation. For this reason, end users will continue to prefer copper-to-the-desk.
Dunnenberger: We think 1000Base-T will serve as a catalyst for FTTD. The intent of the IEEE standards group was to develop a specification to operate GbE over existing, installed Category 5 cabling. While this aim has been achieved technically, the consensus in the cabling industry is that Category 5 is barely adequate.
This has a couple of ramifications: End users are more skeptical than ever of the ability of copper to perform. This is the best explanation for the emergence of Category 5E and proposed Category 6 products. Because most users do not have GbE systems yet, 1000Base-T becomes a technology gamble when planning the infrastructure for future deployment. This skepticism and risk-averseness are leading users toward considering the higher grades of copper cable, which are not standardized and require careful matching of components from a single vendor.
All these factors drive the copper solution to much higher price points while still with no proven track record of success at gigabit speeds. If anything, the ratification of the IEEE 802.3ab standard will signal to the marketplace that GbE-to-the-desk is a technology not of the future but of the present and will spur the search for the best cabling infrastructure solution to support it. This certainly matches the surge of questions, interest, and orders in horizontal fiber from end users that Siecor has addressed over the past six months.
One important trend that warrants closer consideration is the rapid development of voice over IP [Internet protocol]. It is already being deployed successfully in backbone networks-over fiber-and it is just a matter of time before it migrates to the desk. When this happens, a lot will change in network and infrastructure design. Separate voice infrastructure will no longer be required, but the data networks will no longer tolerate sheer brawn, i.e., bandwidth, to deliver high volumes of data packets. Instead, they will need to become very sensitive to latency and quality of service. Lost packets, transmission errors, and retransmissions will not be acceptable. Even greater attention will need to be paid to cabling issues. Fiber's inherent freedom from data-corrupting factors such as crosstalk and EMI/RFI will make it even more attractive as the universal cabling medium of choice.
Jensen: The issue is not that 802.3ab has been ratified. Rather, the issue is the cabling needed to support it. As 1000Base-T is adopted, cabling specifications have already changed from Category 5 to 5E. Fiber is still the best choice to migrate the network, since networking equipment is coming down in price.
Stevens: For the past two years, the copper side of the cabling business has been positioning UTP as a gigabit solution. In that regard, most people have been expecting it and planning on it. Therefore, the release of IEEE 802.3ab will not have a major impact on decision-making. As addressed in the first question of the roundtable, decisions will normally balance performance and cost.
Williams: I don't think 1000Base-T will reduce the market for FTTD. There has been and probably will continue to be a price premium for fiber-to-the-desk versus copper-to-the-desk. The price difference between copper and fiber interfaces for GbE is typical of this long-standing trend. Most customers who implement FTTD are doing so because they have special requirements-for example, security or unusually long cabling distances. For these customers, fiber will continue to be a compelling solution despite the price premium.
Q: Cabling Installation & Maintenance research shows that 23% of end users plan to migrate to GbE within the next few years. What is the single most important piece of advice you would offer them?
Draye: Analyze the requirement of the backbone first and upgrade the backbone (using fiber) to provide the necessary bandwidth to the workgroup GbE switch. It is unlikely that the backbone can support a significant number of GbE horizontal links without having a bandwidth capacity of 10 Gbits/sec. (Before GbE can be deployed to a significant degree in the horizontal, 10-Gbit/sec development for the backbone may have to become an economic reality.)
Dunnenberger: It is time to take another look at fiber as a part of the solution for your horizontal cabling. As copper solutions become ever more expensive, fiber solutions are becoming more affordable. The advent of the MT-RJ small-form-factor connector has had a profound impact on both cabling infrastructure costs as well as the cost of network electronics components from many major vendors.
Jensen: Ensure that you're ready with the right cabling solution. Evaluate all of the solutions from a technical perspective. Some of the things to think about with copper cabling solutions are the electrical performance specifications (jacks, cable, and patch cords); the temperature variations in which the cable will be placed (copper cabling attenuation increases in higher temperatures, so don't place it in attics); alien crosstalk (don't place your cables next to each other-for instance, in a furniture system); EMI (don't place it next to power cabling), and evolving cabling standards.
For fiber, you need to think about placing either 50/125- or 62.5/125-micron fiber. Attenuation and length of cable are your only cabling concerns. The bandwidth of the cable may be a factor once you start looking at 10-GbE-to-the-desk.
Stevens: My advice is to look beyond the 1-Gbit platform. Just as 100 Mbits/sec was a transitory step, so, too, is GbE. Work has already started on 2.5 and 10-Gbit solutions, so consider a cabling plant that can best address the next level. In that regard, always install a quality, high-performance cabling infrastructure. The few more dollars invested today will pay huge dividends when it is time to migrate to the next level. For fiber, put in the highest bandwidth multimode available. Overbuild backbones with singlemode fiber; it will be used. Use Category 6 UTP because it will offer the best migration path to technologies beyond 1 Gbit.
Williams: The quality of your cabling will matter more than it ever has in the past. So it is important to properly specify cabling in new facilities and ensure that it is properly installed and certified.