As Gigabit Ethernet becomes a reality, installers must learn about the “active” side of testing.
Although not often acknowledged, there is a great divide in the world of network testing. On one side are cabling installers, who perform acceptance testing before turning a new installation or network upgrade over to their customers. On the other are network managers who expect their installations to meet the needs of their clientele, the users of the network.
The unmentioned reality here is that the acceptance tests performed by cabling installers are designed to measure the electrical and optical parameters of installed cable, connectors, and other passive components. Network managers, on the other hand, depend on different tests, using different test equipment, to measure different parameters. The bottom line is network managers aren't interested in the integrity and performance of the physical layer comprising cable and connectors. Their main concern is performance.
A great divide results from the fact that the electrical and optical parameters the cabling installer tests are only theoretically connected to active network performance. However, that divide has begun to be surveyed, and even crossed, within the last year — thanks to the appearance of Gigabit Ethernet.Gigabit Ethernet changes everything. The world of voice/data/video cabling is changing rapidly. Five years ago, Gigabit Ethernet (GbE) was only a distant cloud on the technological horizon, and 100-megabit/sec Ethernet (labeled Fast Ethernet to distinguish it from 10-megabit/sec Ethernet) was all the rage. Today, GbE is here, and 10-gigabit/sec Ethernet is well into the planning stage.
The catch in all this is that GbE, operating at 1,000 megabits/sec, represents a tenfold increase in capacity over what top-of-the-line cabling infrastructures were designed to handle. Until recently, most considered a Cat. 5 cabling system the acme in performance, with more bandwidth than anyone was likely to need for decades. Cat. 5 cabling systems, however, are tested to only 100 MHz.
Can ordinary Cat. 5 cabling infrastructures carry GbE? This debate has been raging for two years. The problem is Cat. 5 cabling systems are designed to operate over only two pairs of a 4-pair cable — one pair serving the transmitter and one pair the receiver.
To move gigabits of information over a medium designed to carry megabits, GbE uses all four pairs of the cable and requires bidirectional transmission, with a transmitter and receiver transmitting on each pair. If you do the math, you'll see that each pair must still carry 250 megabits/sec on a cable designed for 100 megabits/sec. The cable makes up the difference by using sophisticated data-encoding schemes designed by electronic engineers to place multiple bits of data on a single hertz of bandwidth (see Sidebar below).
Using all parts of a 4-pair cable for bidirectional transmission as opposed to using two pairs for transmission in one direction places far higher demands on the cable. The debate over whether or not ordinary Cat. 5 cable can carry GbE stems from the fact that both the quality manufactured into such cables and the expertise and craftsmanship with which they are installed have consistently improved over the past five years. The obvious conclusion is some Cat. 5 installations can carry GbE, but the only way to find out if yours is one of them is to test it (see Table 1 above).Escalating category race. Manufacturers of Cat. 5 cable recognized early on that there would soon be a need for bandwidth greater than 100 MHz. Many of them began offering Cat. 5 labeled “enhanced” or “extended-performance.” Major voice/data distributors, such as Anixter with its Levels program and Graybar with its VIP program, also established higher performance standards than the Cat. 5 minimum identified in the ANSI/TIA/EIA-568 series of cabling standards.
Soon the Telecommunications Industry Association (TIA) followed suit with a subcategory of Cat. 5, labeled Cat. 5E. Cat. 5E cable has to pass additional tests for electrical performance specifically designed to accommodate GbE. For example, a near-end crosstalk (NEXT) test for Cat. 5 cable measures the crosstalk disturbance on a wire pair caused by each of the other three pairs of the cable in turn, and establishes whether or not the cable passes or fails the test on the basis of the electrical performance of the worst pair-to-pair combination. This is an appropriate test for systems operating over two pairs, as standard Cat. 5 systems do.
However, GbE systems require a different testing procedure because they use all four pairs bidirectionally. The appropriate approach for GbE tests is power-sum testing, which measures the total crosstalk effect of three transmitting pairs on the remaining pair in a 4-pair cable. In addition to power-sum NEXT (PS-NEXT), Cat. 5E networks must also pass other tests not required of Cat. 5 networks: equal-level, far-end crosstalk (ELFEXT); power-sum ELFEXT (PS-ELFEXT); and return loss (Fig. 1 on page 80).
As a result, most industry analysts recommend a minimum of Cat. 5E cable and components to accommodate GbE. Some go so far as to recommend a minimum of Cat. 6 cable, which will be tested to operate to 250 MHz of bandwidth. (Many manufacturers already offer Cat. 6 cable, even though the TIA has not yet finalized a performance standard for it. Current Cat. 6 offerings are based on provisional performance requirements that may change, manufacturer guarantees that products will comply to the final specification notwithstanding.)Enter active testing. So far, we've only addressed the cabling infrastructure side of the great divide. What about active network testing? There are several compelling reasons why cabling installers have not been involved in this function in the past. For one thing, the cabling installer is usually working at a construction site, especially in the case of new construction. The building may not be occupied, the network equipment is unlikely to be in place, and there may not even be electrical power. An installer may go months without meeting a network manager or having to work on an operating network.
All of that may be changing, however, as a panel discussion at last fall's BICSI conference illustrated. The presentation, “Testing to Real-World Applications to Assure Network Quality of Service,” compared active and passive network testing.
Presenter Daniel Kennefick, business manager of cable manufacturer Berk-Tek, New Holland, Pa., pointed out that “real-world network managers typically need reliability, improved performance, increased bandwidth, and low maintenance. What this really means is that they need their networks to perform like utilities, and the key to utility-like performance is increased performance margin in the cabling infrastructure.”
According to Kennefick, installation problems such as faulty cabling and improper installation and termination practices are the second most likely causes of Ethernet errors after faulty active equipment. The kinds of network errors caused by these problems include CRC, or bit errors, late collisions, excessive collisions, alignment errors, runt packets, and fragments.
“Among the best ways of solving these problems,” Kennefick said, “are correcting improper termination and installation practices, as well as improving key cable performance characteristics such as return loss, crosstalk, and attenuation.”
He added that test evidence has shown a generic Cat. 5 cabling system can have a bit error rate (BER) of about five times the minimum Ethernet requirement of IEEE, while a Cat. 6 cabling system tested at the same time had a BER of only 1/100 the IEEE minimum.Physical layer is the key to throughput. Fanny Mlinarsky, general manager of the WireScope Operation of Agilent Technologies, Palo Alto, Calif., had a similar take on network performance. She suggested today's bandwidth-hungry business applications — such as informational broadcasts, distance learning, video telephony and conferencing, and tele-medicine — demand a low BER, and “it is the physical layer that determines BER, and therefore data throughput.”
High-quality transceivers, the right kind of cable, and proper network configuration and operation are also important factors in quality of network service, but, Mlinarsky concluded, “the integrity of the physical layer is the foundation and key to good throughput performance.”
Tim Takala, an industrial engineer working for Krone, Englewood, Colo., pointed out that laboratory tests are typically conducted on 94-m cable channels or 100-m cable links, whereas cable runs in the real world were often much shorter. Also, field testing typically takes place without patch cords in place — adding patch cords often changes everything.
“What is the ‘fitness for purpose’ of a structured cabling system?” Takala asked. “I've got a news flash for you. The network manager doesn't care if a jack has three decibels of headroom over the Cat. 6 specification for NEXT. It's not compliance with TIA requirements, but transporting data reliably from A to B — that's what the network manager cares about.”
Takala defined fitness for purpose as a network that satisfies its end-users, both now and in the future, and does so within budget and on schedule.
“Network efficiency is affected by two things: downtime and slowtime,” he said. “Outages cause 72% of the problems on a network, and service degradations about 28%. Typically the end-user only notices outages and takes slowdowns in stride, but both problems affect business productivity.”
Takala had several suggestions for fighting the major causes of network downtime and slowtime. “Manufacturers should test their products to real-life installation practices, including realistic run length,” he said. “Also, cabling installers should test channels vs. links. And networks should be tested actively after installation.”
Takala added that tuning the network for greater performance adds an average of 3% to a company's bottom line due to productivity increases.Improving network performance. This leaves us with several ways to improve active network performance:
Install the highest category of twisted-pair cable and passive components you can afford, consistent with approved TIA standards. A robust cabling infrastructure can overcome some active network problems.
Make sure trained craftspeople install your network. Controlled testing has shown that such improper installation factors as excessive pair untwist and pulling tension can affect electrical performance.
Test installed cabling infrastructure as realistically as possible, including active testing with network equipment in place where it is feasible.
Tune your network, keeping in mind that many active-network problems result from or are influenced by installation and termination practices, as well as the overall quality of cable, connectors, and other passive components.
Powell is an editorial consultant to EC&M. Please send comments or questions to [email protected].
Sidebar: Megabits vs. Megahertz
A common misunderstanding among cabling installers is that “megabits” and “megahertz” are interchangeable terms. Fostering the misconception that these two terms refer to essentially the same thing is the fact that they are often equivalent — in a certain sense.
To put it simply, the hertz is a measure of bandwidth, the size of the pipe through which signals are sent. The bit is a measure of throughput, or the speed with which things move through the pipe. In some cases, a million hertz (or megahertz) of bandwidth carries a million bits (or megabit) of information. This occurs when one bit of information is transported by one hertz of bandwidth, represented as one cycle of a sine wave.
However, as the need for throughput has increased faster than provision of bandwidth, electronic engineers have developed a series of ingenious encoding schemes for putting more and more bits of information on a single hertz of bandwidth. For example, an encoding scheme with a 4:1 ratio of bits to hertz would allow four bits of data to be transmitted on one hertz of bandwidth. Theoretically, then, 400 megabits/sec would be the maximum data rate over standard Cat. 5 cable using a 4:1 encoding scheme.
Less common — but still erroneous — is the confusion between bits/sec and the term “baud.” Technically, the baud rate is the rate at which a signal changes from one electrical state to another on a transmission line, while the rate of flow of binary digits, or bits, over a line is measured in bits/sec. Again, in simple systems without encoding, the two rates may be equal, but the terms are not equivalent.