Technology experts keep building the case for standards groups to pick up the initiative.
by Patrick McLaughlin
Senior vice president of research at Nvidia (www.nvidia.com) Bill Dally caused quite a stir throughout the computing industry in April 2010 when in a commentary for Forbes he declared Moore's Law dead. In that column he said central processing unit (CPU) "performance no longer doubles every 18 months." While making the argument in his column, Dally pointed to the infrequently referenced other prediction made by Gordon Moore–that in addition to the number of transistors on an integrated circuit doubling every 18 months, "the amount of energy consumed by each unit of computing would decrease as the number of transistors increased. This enabled computing performance to scale up while the electrical power consumed remained constant. This power scaling, in addition to CPU scaling, is needed to scale CPU performance.
"But in a development that's been largely overlooked, this power scaling has ended," Dally wrote. The end of that power-scaling was the death knell for Moore's Law, he argued.
Moore's Law has been the foundation of computing-based industries for decades, and certainly the networking industry supported by structured cabling systems is among them. During a seminar produced by Cabling Installation & Maintenance and delivered via the Web in January 2011, Moore's Law and power consumption were part of a dialogue about the capabilities of shielded twisted-pair cabling systems.
Standard development
Two of the speakers who participated in that seminar were David Hess, technical manager for standardization and technology with Berk-Tek, a Nexans company (www.berktek.com) and Valerie Maguire, global sales engineer with Siemon (www.siemon.com). Hess and Maguire have participated in a number of groups, some of them formal standards-making bodies and others less-formal groups made up of networking and cabling technology experts, which have explored the possibilities for shielded twisted-pair structured cabling systems to support Ethernet at transmission speeds higher than 10 Gbits/sec. Hess's presentation focused on the standards scene and what is happening in the realm of standards activity for such high-data-rate cabling systems. Maguire's presentation dealt with the "hows and whys" of a shielded system's ability to support such high speeds.
In setting the landscape for current activity, Hess explained that twisted-pair Ethernet development has continued for more than 20 years. 10Base-T specified and referenced telephone inside wiring, then during the development of 100Base-TX the decision was made to articulate the cabling and formalize the reference. He then noted that, "Cabling and PHYs are essentially developed in parallel. Each new category [cabling] definition begins from an extrapolation of the last category. And co-development of cabling and PHY standards must be coordinated."
Then Moore's Law made its way into the conversation. Hess presented a chart that plotted the Base-T specifications developed by the Institute of Electrical and Electronics Engineers (IEEE; www.ieee.org) and the Category cabling standards developed by the Telecommunications Industry Association (TIA; www.tia.org) as well as the International Organization for Standardization (ISO; www.iso.ch), based on the years the standards were approved by their respective organizations. Also on the chart was a trend line that equated to Moore's Law. The points representing IEEE "Base-T" specs and TIA/ISO "Category" specs plotted right along the line representing Moore's Law up to 2000, where 1000Base-T and Category 5e reside. "More recent standards have taken longer to complete," Hess stated. The 2006 10GBase-T specification was a year behind the Moore's-Law line that crossed the 10,000 Mbit/sec data-speed point at 2005. The Category 6A data point was even a little later than 10GBase-T.
The next point made by Hess, which Maguire later reinforced and explained in detail, bears a striking similarity to the sentiments Nvidia's Dally expressed in his Moore's Law death declaration. "Two dates are given for 10GBase-T," Hess said. "2006 for 802.3an, the primary PHY approval, and 2010 for 802.3az, Energy Efficient Ethernet PHY approval." Energy Efficient Ethernet enables lower-cost 10GBase-T operation, he explained.
Power in focus
Dally said the reason Moore's Law is dead is because it is no longer true that each computing unit consumes less power as the number of transistors on an integrated circuit increases. It would be far from conclusive to point to the delayed development of Base-T Ethernet and Category cabling specs since 2000 as proof that Dally has it right. However, Maguire voiced the opinion during the January seminar that power consumption will be a significant consideration for those setting their sights on twisted-pair-based applications at speeds beyond 10 Gbits/sec.
Interestingly, the cause-and-effect sequence that eventually leads to an examination of power consumption begins with a reference to Moore's Law. "In the data center, server-cluster storage rates are increasing faster than Moore's Law predicts," Maguire said. "They are doubling every year as opposed to every 18 months. This tells us that the need for high-speed throughput in the data center is growing at a faster rate than for high-speed twisted pair applications for Ethernet in the LAN. We can predict that in three to five years we are going to have a need to support 40-Gbit/sec Ethernet in the data center."
In corporate LAN environments, she explained, growth is being driven by video applications and high-performance computing. Transmission-rate increases are not as rapid in the corporate LAN as they are in the data center. Maguire said she expects LANs to begin having a need for 40-Gbit/sec Ethernet in five to ten years.
As for data centers' three- to five-year timeframe, "We need to start planning for that application now," she said, and that is why companies such as Siemon and Nexans have shared ideas and participated in industry activities to further investigate the practicalities of higher-than-10G over twisted-pair.
"Data center experts are telling us they're very concerned about two issues: cooling and power. We need to take this consideration into account when we plan for the capability of the beyond-10GBase-T application. We need to make sure that powering is not an issue, and heat generation is not an issue. We need to take into consideration this feedback from data center experts, and ensure the technology we use to support greater-than-10GBase-T isn't power-hungry and that it supports the optimum throughput capability of the media."
As of January 2011 there was no official Project Call for Interest within IEEE for a greater-than-10G application over twisted-pair. Without such a call, no one can make application-support claims. Maguire explained why. "There are a whole range of considerations that would affect the capability of a 40-Gbit/sec application over twisted-pair. These include the level of encoding, the level of cancellation and strategies to alleviate power consumption." With no call for interest there are no agreed-upon models for considerations such as these. "We can generate our own capacity models," Maguire explained. "We can look at channel-length histograms for data center environments. And we can share our expertise among companies that are interested in pursuing this initiative to come up with our best guess of what the media and the topology to support a greater-than-10-Gbit/sec application would look like."
While no official activity is taking place within the IEEE, Berk-Tek's Hess noted, "The trend toward parallel development is continuing. And interest is forming among 10GBase-T solution providers."
The crystal ball
Maguire engaged in some crystal-ball gazing, offering her own vision of what the realm of beyond-10G-over-twisted-pair will look like. "My belief," she said, "is that we're going to break this project up and look at the immediate need of the data center, and reserve research into the needs of the LAN environment for later in the project or in a future project.
"I think it's going to be extremely desirable not to significantly increase the equipment complexity beyond existing 10GBase-T levels. This means, if anything, we want to reduce cancellation. We want to reduce power consumption. We do not want a more-sophisticated or more power-hungry chip technology."
The cancellation of noise might be viewed as a central issue that ties together several elements discussed so far in this article: Moore's Law, power consumption, increasing data rates for Base-T Ethernet, and shielded cabling systems as a potential solution. As Base-T standards have gone from 10 to 100 to 1000 to 10,000 Mbits/sec, the frequency range over which data signals are carried has also increased. 10Base-T uses the frequency range up to 20 MHz, whereas 100Base-TX uses the frequency range up to 31.25 MHz, 1000Base-T uses twice that frequency range, up to 62.5 MHz. For 10GBase-T, there is a substantial increase to 413 MHz, but the increase is still not 10x. Efficiency is also being increased for each new generation.
Hess pointed out that while speeds increased by a factor of 10 with each successive specification set, "the most outstanding measure of progress is evidenced in the efficiency, that is, the bits-per-second-per-hertz." By that measure, computing efficiency went from 0.5 bits/sec/Hz with 10Base-T, to 3.2 with 100Base-TX, to 16 with 1000Base-T, and to 24.2 with 10GBase-T. Hess added, "There is significant further study needed to consider how far 10GBase-T complexity might be surpassed for the next generation."
Quieting the noise
Even with such advancements in efficiency, it remains true that as transmission-frequency increases, more noise is encountered and therefore, noise cancellation has become a more-complicated and more-challenging task for creators of Base-T standards. And once noise-cancellation is defined, its very existence in a transmission specification has side effects. Maguire explained, "The 10GBase-T application uses quite a bit of digital signal processing [DSP] to provide crosstalk cancellation. A result of this DSP is an increase in latency. And latency can be very critical for applications."
In addition to developing ways of cancelling noise, effort goes into strengthening a signal's transmission to begin with. The efforts are complementary methods of ensuring a receiver can discern between the transmitted signal and the noise. Signal-strengthening has its consequences in the realm of chip design. Maguire described a wish list for the design of a chip for greater-than-10G on twisted-pair, with attention on reducing power consumption and latency: "I'd like to reduce a couple items in chip design. One is line-driver power. I'd also like to reduce the number of chokes and transformers in my system. If I consume less power I'll have a solution that requires less tooling. If I'm doing less digital signal processing I'll have a solution that has better latency performance. That will satisfy the needs of my customers."
The complexities of chip design, power consumption, digital-signal processing, noise cancellation and other electro engineering issues are what lead Hess, Maguire and others in the industry to consider shielded cabling systems as the medium best suited for the next-generation Base-T specification, if and when one is created. On the reduction of line-driver power, Maguire noted, "That can be done in a couple ways. One is to use a media type that has improved insertion-loss performance, such as Category 7A. I can also accomplish it by deploying over a shorter channel length. The optimum performance will be realized by doing both." Reducing the number of chokes and transformers can be accommodated "by using a shielded cable that has superior electromagnetic immunity performance," she added.
But it's not as if a Base-T solution is the only option for Ethernet transmission at speeds greater than 10 Gbits/sec. Optical and twinaxial versions of 40-Gbit/sec Ethernet already exist. Why would or should a standards body go through the effort to produce another option? "It's related to some shortcomings with the existing twinax and optical-fiber solutions, as well as one of the features of a Base-T application," Maguire opined. "The high-speed twinax solution supports short-length topologies, less than seven meters," she said. "This means the configuration is ideally suited for top-of-rack and possibly some middle-of-row deployments. But it does not support a structured cabling topology, which is required by the TIA-942 and pending TIA-942-A standards."
The multimode-fiber versions of 40- and 100-Gbit Ethernet "require serial transmission that uses 8- and 20-fiber links," Maguire explained. "The effect is the user may be losing some density because, in order to support these 8- and 20-fiber links, you need to use MPO-style cassettes, which do not have the density to this point of LC connectors."
She then pointed out a characteristic that distinguishes Base-T applications from others. "These are the only applications that support autonegotiation, which means it can match speeds between servers and switches. This is very important because deploying Base-T-type applications means when it's time for a network upgrade, you do not have to upgrade your servers and switches at the same time. So there is a real need for a twisted-pair beyond-10GBase-T application.
Even with no formal activity taking place within the IEEE, Maguire made known her own perspective on how things could and may play out. "I believe the next application speed is going to be 40-Gbit/sec, and I believe that would be agreed upon by most industry experts. The longest support or reach of a 40-Gbit over twisted-pair application would be over Category 7A cabling. That doesn't preclude the use of Category 6A, but Category 7A will support the longest reach and most flexibility in the data center. I also think that for the initial, data center 40-Gbit/sec solution we might be looking at a reduced-length topology." As mentioned earlier, Maguire foresees a data-center-focused specification being developed first, followed by a specification more geared toward a corporate LAN environment.
She was up front about the fact that advocacy of Category 7A is not in keeping with the way the IEEE has approached Base-T standards to this point. "This is going to require a change in thinking among IEEE experts who develop Ethernet standards," she said. "Historically IEEE has targeted cabling that has the largest installed base. But in this case, we have to question whether that choice makes sense–if we choose the highest-performing grade of cabling versus the most widely installed Category of cabling. We also need to look at the topology. Traditionally we have targeted the 100-meter, 4-connector topology. But in this case if we are looking for an application that is targeted to the data center, doesn't it make sense to look at a topology that matches the needs of the data center?"
She then pointed out that end-user organizations can benefit from using the highest grade of cabling. "The benefit is extreme and is realized in the form of cost savings," she said. In a typical IT market expenditure, 17 percent is represented by IT hardware, which includes cabling. Other expenditures including software (18 percent), salary and benefits (36 percent), IT services (13 percent) research-and-development (5 percent), and others (11 percent).
"What's interesting is if I further break down the hardware into more detail, cabling is typically 2 to 3 percent of the total hardware spend," she noted. "So I have two options. I can install a higher grade of cabling that doesn't represent most of the installed base but represents an additional 2 or 3 percent of the IT spend. Or I can develop equipment that has very high power needs, perhaps increased latency, and costs significantly more, potentially increasing the overall percentage of the hardware to be significantly beyond 17 percent. The benefits of specifying a superior media type is cost savings in terms of the network electronics we're going to deploy to support 40GBase-T and beyond."
Hess and Maguire made a compelling case to A) move forward with 40GBase-T, and B) recognize shielded twisted-pair cabling as the preferred medium for 10GBase-T. How compelling it is to other industry experts who would participate in such specification creation will be interesting to track. We will continue to cover this topic, and plan to provide you with information on any related standards activities that take place..
Patrick McLaughlin is chief editor of Cabling Installation & Maintenance. The Web seminar from which this article's quotes were taken can be viewed at cablinginstall.com.
Past CIM Issues